The studies we depend on, reproduced

Companion to [Questions we need Emily to answer](/writing/questions-for-emily-testing-a-theory-of-the-interface) and [the essay](/writing/testing-a-theory-of-the-interface-with-an-ontology). Generated by `scripts/reproduce_studies.py`.

Each page takes one paper we depend on, writes its sample and process history in our language as the paper states them, simulates it, and sets the result beside the paper's numbers. The sample and history were re-read from each PDF by a second agent that had not seen our transcriptions, and every item is marked: same, approximated, assumed, differs, or not in ours. Nothing was fitted for these pages; every model runs with the parameters recorded when it was last scored. No value has been checked by a person, which is what these pages are for.

Two different grades. Reproducible from the paper? grades the paper as a source: whether it states its sample, every step and every step's timing well enough that a simulation needs no choices of ours. It changes only when more is learned about the experiment. How well our simulation matches grades us: the share of the paper's numbers our simulation lands within tolerance of (High: three quarters or more; Moderate: two fifths or more; Low: fewer), with the median error, what the alignment adds on samples held out of it, and the first score recorded for the study. This is the column that moves as the language and the programs improve.

How to read a page

• The player shows the sample in three dimensions with its history on a clock. Layer thicknesses and read-outs are the simulation's values at that moment.
• The match diagram puts the paper's words on the left and what we simulate on the right, one row per sample property and per step.
• Their terms and ours maps the paper's vocabulary onto the language and the ontology.
• The simulation in our language is the complete input, and the model program as written.
• The difference chart shows, for each number the paper reports, how far our simulation is from it.

What the second reading found, across all studies

• Vitkavage, Irene and Massoud 1990. Every wafer was pulled in under 5 s in the oxidizing gas. The recipe stored with each measurement stops at the end of the oxidation (plus the anneal); the pull is on the source record, not in the history. Written as a step it moves the simulated fixed charge of the (100), 1000 °C, 1 h wafer from 1.46 to 1.58 × 10^11 cm^-2.
• Vitkavage, Irene and Massoud 1990. The report describes two sample sets. Dry oxygen under 6 ppm water, the HF-last clean and boron doping are stated for the non-cleanroom set only; for the cleanroom set, which all our electrical values come from, the gas is never named. Our library records one set of conditions for the source.
• Deal, Sklar, Grove and Snow 1967. The recipe stored with each of these measurements is one step: the pull, or the anneal. The oxidation that came before it (1200 °C, 60 min, dry oxygen, 0.2 µm) is absent, and so is the pull that must follow an anneal.
• Deal, Sklar, Grove and Snow 1967. The argon anneal at 920 °C is stored as inert_N2. The language has had inert_Ar since the SiC work; the row was never updated.
• Deal and Razouk 1980. The stored recipes begin after the oxidation. Written out in full, each history is the oxidation (300, 84 or 28 min), the anneal if any, the pull, and ten minutes at 400 °C after metallization.
• Deal and Razouk 1980. The argon runs (TR-4, 8, 12, 16, 20, 24) are stored as nitrogen anneals. The paper treats the two gases as one curve, so the numbers are not affected; the record is wrong.
• Hung 1987. The stored recipes have one or two RUNs: the oxidation and, in chapter 4 and the anneal series, the nitrogen anneal. The thesis's Appendix adds a 3 min boat-in in the oxidation gas, a 3 min boat-out in nitrogen (chapter 4), a fast pull in the oxidation gas (chapter 5), and in chapter 4 a 30 min forming-gas anneal with 15 min of nitrogen before and after. None of these is in the stored recipes.
• Hung 1987. The anneal series P (1050 °C, 20% O2) is stored with 27 min of oxidation, “inferred from the thickness”; Table 5.1b prints 20 min. The datasets were built before Table 5.1b was read.
• Song, Dhar, Feldman, Chung and Williams 2004. Our stored recipe (one hold in dry oxygen at the stated temperature) is what the paper states. The second reading found nothing our encoding contradicts.
• Song, Dhar, Feldman, Chung and Williams 2004. The paper never says whether the points on a curve are separate samples or one sample re-oxidized. Our simulation treats each time as an uninterrupted oxidation from a clean surface.
• Kobayashi et al. 2017. Oxidation and argon anneal were done in different furnaces, so each annealed sample was cooled, unloaded and reheated in between. Our stored history runs the anneal straight after the oxidation.
• Kobayashi et al. 2017. Five minutes of argon and forty minutes of oxidation at 1300 °C are our choices from the ranges the paper gives (1 to 10 min; 40 min or 1 h 20 min). The stored records present them as the recipe.
• Goncharova et al. 2011. The stored recipe gives the anneal as dry oxygen at one atmosphere. The paper's gas is 18O2 at 0.01 Torr, about 1.3 × 10^-5 atm. No program reads the pressure, so the numbers do not change, but the fitted rates belong to that pressure and cannot be carried to another.
• Goncharova et al. 2011. Our oxygen dataset labels the lowest temperature 773 K and our profile dataset labels the same anneal 763 K. The paper uses both, and 490 °C; the figures say 763 K.
• Akinwande, Ho and Plummer 1982. The stored recipes (data/derived/akinwande_1982_nf.csv through scripts/canonical_measurements.py) begin with the 2 s pull. The 60 min of argon pre-heat, HCl etch and purge and the 75 min oxidation are not steps; the oxidation enters the program only as a constant (T_ox = 1000 °C). For the charge this changes nothing, because the program starts from the steady value at the oxidation temperature.
• Akinwande, Ho and Plummer 1982. Every argon anneal is stored as AMBIENT inert_N2. The report says argon with under 1 ppm water. Both have no oxygen, so the numbers are the same; the record is wrong.
• Razouk and Deal 1979. There is no dataset in data/derived for this paper and no stored recipe: the project's analyses read figures_agent_read_UNVERIFIED.csv directly, with the oxidation times 360 and 60 min for Figure 7 only and no history for Figures 8 and 13.
• Razouk and Deal 1979. The two readings of the figures (ours earlier, and the second reading) agree within 5% on 41 of the 44 values both hold. They disagree on three: Figure 13 after hydrogen at 1200 °C (0.60 earlier, 0.52 in the second reading, 13%), Figure 7 (100) at 1200 °C, 10 min in nitrogen, unannealed (3.2 against 3.0, 6%) and the same in argon for 60 min after hydrogen (0.123 against 0.13, 6%). The values compared on this page are the earlier reading's, which the project's analyses use. The second reading also gives what the first did not: Figure 13 before hydrogen and after a 450 °C, 25% H2 anneal, and the fixed charge of Figures 7 and 9.
• Deal and Grove 1965. Our transcription of Table II (tables_agent_read_UNVERIFIED.csv) and the second reading agree in all 24 cells: none differs.
• Deal and Grove 1965. The program does not carry the paper's activation energies. It carries a least-squares Arrhenius line through the six Table II rows: E_B = 1.286 eV against the paper's 28.5 kcal/mole (1.236 eV), and E_BA = 2.030 eV against 46.0 kcal/mole (1.995 eV). At the table's own temperatures the line is off the table by up to 15% (B at 1200 °C, B/A at 800 °C); see the table below.
• Krzeminski et al. 2007. The stored recipe for the paper's own wafers is one RUN of 90 min at 725 or 750 °C in dry oxygen. The paper's cycle is a load at 675/700 °C, a 30 min ramp in 9% oxygen, the main stage in oxygen, 10 min of ramp-down and 30 min of reflow in nitrogen. The ramp is the one that matters: it grows 1.53 or 1.96 nm, all of the oxide at “0 min”.
• Krzeminski et al. 2007. The other-laboratory score in evaluations/growth_search_report.json (6 of 10, 13% low at 725 °C) started the law at the measured 1.53 or 1.96 nm. The predictor cannot do that: it starts every history at the fitted 1.71 nm. Run through the predictor with the stored recipe, the same points come out between 1% high and 14% low at 725 °C and 8 to 11% low at 750 °C, and the 0 min points (excluded from the search's score) 12% high and 13% low. The two scores of one program on one dataset differ because the starting oxide lives in different places.
• Müller, Lecestre, Demoulin, Cristiano, Hartmann and Larrieu 2023. The stored recipe for every Si wire is one steam step of 779.5 s at 850 °C. That is the time the bulk slope of Figure A1 (1.46 nm/min) needs to grow 19 nm from zero; the paper's example time is 7.7 min (462 s). The difference, about 5.3 min, stands in for whatever oxide the bulk had before the steam: the ramp in oxygen, the 5 min hold, the native oxide. With 462 s of steam and nothing before it, the program gives the 94 nm wire 21.8 nm of oxide instead of the measured 33.4.
• Müller, Lecestre, Demoulin, Cristiano, Hartmann and Larrieu 2023. The paper's furnace cycle has five simulated steps (nitrogen ramp, oxygen ramp, 5 min oxygen hold, steam, nitrogen cool-down); the stored recipe has one. The radial program grows nothing in the four we added, because its rate is proportional to p_H2O and it has no temperature dependence.
• Krzeminski, Han and Larrieu 2012. The stored recipe is one steam step of 600 or 1200 s at 850 °C. The paper states 850 °C, “wet” and the two times, and nothing else; the stored recipe adds pure steam at 1 atm. Our history adds an assumed ramp and an assumed oxide at time zero.
• Krzeminski, Han and Larrieu 2012. Every curve of Figure 1(a) starts at about 5 nm at t = 0. The radial program starts from bare silicon, so either the program misses 5 nm or the paper's times do not start at the start of oxidation. Writing that oxide as 90 s of steam before the stated time makes every wire thicker by 1 to 5 nm, which is the wrong direction for this paper.
• Rabie, Haddara and Carette 2005. The stored LeGoues recipe is one step, 180 min of wet_O2 at 800 °C. wet_O2 carries no partial pressures, so the inventory program runs it as pure steam (p_H2O = 1). Written as the 55% steam that the paper's own C* implies, the inventory's enriched layer falls from 91 to 69 nm and the alloy consumed from 65 to 49 nm; the Rabie program's thickness does not change because it does not read the gas.
• Rabie, Haddara and Carette 2005. The germanium-rich layer (50 nm, up to 75%) has no time in the paper. Our dataset row puts it at 180 min, the last point of Figure 4, and scores it as a value (75%) although the paper says “up to”.
• David et al. 2015. The dataset holds two rows of this paper (50% at 800 °C, 90 min; 50% at 750 °C, 6 h). The text supports the same 50% at 750, 850 and 900 °C too, but Figure 3b peaks near 45% at 900 °C, so the 900 °C row is the paper's sentence against its own figure.
• David et al. 2015. The stored recipe of the 6 h row is one 6 h hold. The paper's sample is a 34 nm film on 12 nm of silicon on oxide; by 6 h all the alloy is consumed. No stored field says so.
• Wang, Hu and Irene 1994. The stored recipe for each temperature is one hold in dry oxygen for the longest time. The paper's sequence has a heat-up in vacuum, a fill, the oxidation, an evacuation and a cooling in vacuum; written out, the heat-up and the cooling add about 0.04 nm in the program because it does not read the gas.
• Wang, Hu and Irene 1994. The starting oxide is a fitted parameter of the program (“starting oxide fixed” in the selected structure, a value fitted with the others). The paper's two figures disagree on it (about 0 nm in Figure 1a, about 1 nm in Figure 1b); the sample cannot carry it.
• Kao, McVittie, Nix and Saraswat 1988. The project held no dataset for this paper (ontology/library.json: “Not modelled”). The flat-oxide table was in the history bundle's EXTRACTED_DATA.json (Part I Table 1) and agrees with the second reading; the curved-surface figures had not been digitized. Every curved number here is a new graph reading.
• Kao, McVittie, Nix and Saraswat 1988. The radial program reads RUN_FLAT_OXIDE_NM as oxide ÷ steam time, a linear rate, and then divides by its diffusion term 1 + (2a/A) log(b/a) with A = 1.1 µm. For Kao's 460 to 555 nm flat oxides the program's own flat sidewall then grows only 70% of the flat oxide it was given (349 nm for 503 nm at 800 °C). Kao's normalization needs the curved oxide over the flat oxide of the same program.
• Kao, McVittie, Nix and Saraswat 1987 (Part I, report). The project held Table 1 in the history bundle's EXTRACTED_DATA.json (marked “transcribed and visually checked”, not checked by us); it agrees with the second reading digit for digit. No dataset in data/derived uses it, and the figures had not been digitized.
• Kao, McVittie, Nix and Saraswat 1987 (Part I, report). The planar wet program's constants are Deal and Grove's at 640 Torr of water from 920 to 1200 °C. Kao's 800 and 900 °C runs are outside that range, and the steam source is not stated: pure steam at 1 atm is 19% faster than the bubbler the constants were measured with.
• Raider and Flitsch 1978. The project had no dataset, no stored recipe and no transcription for this paper; the library marks it “skimmed”. Every number on this page is from the second reading.
• Raider and Flitsch 1978. SOURCE.md and the library note say the region “depends on orientation, not on other process parameters”, which is the paper's summary. The second reading adds what that rests on: one fitted width per orientation in 0.25 nm steps, from a cloud of films whose recipes are not individually given.
• Johnson, Biegelsen and Moyer 1981. The project had no stored recipe for this paper: five sentences in data/primary_sources/johnson_biegelsen_moyer_1981_jvst_19_390/text_values_agent_read_UNVERIFIED.csv, no dataset, no script. Those five values agree with this reading.
• Johnson, Biegelsen and Moyer 1981. That file reads “onset of spin-centre loss with aluminium … 250 C”; Fig. 2 shows the with-Al density still at 1.0 at 250 °C and falling between 275 and 350 °C. 250 °C is where the fall begins, not where it is half done.
• Stesmans and Afanas'ev 1996. The project had no stored recipe and no dataset for this paper; SOURCE.md summarises the text and says Figure 1 was not digitized. The values in SOURCE.md agree with this reading.
• Stesmans and Afanas'ev 1996. SOURCE.md says “Up to about 640 C the density recovers to the as-oxidized pool, 4.7e12”. The 4.7×10¹² is the authors' fit (N0) under the HPb-dissociation model, and the paper calls it “the natural Pb bath … found in the as-oxidized state” (p. L508); no as-oxidized measurement before passivation is shown.
• Yano et al. 2002. The project held this paper as “held, not yet transcribed”: there was no stored recipe and no dataset. The Dit values here are the second reader's digitization of Figure 3 (pixel calibration), not a project dataset.
• Yano et al. 2002. The project's note calls the paper “interface state density against energy on two faces … (and the effect of anneals)”. The paper has one process for both faces and no anneal comparison: one wet oxidation, one argon anneal and one forming-gas anneal for every sample.
• Gupta and Akhtar 2011. The project held this chapter as “skimmed”, with the tables transcribed in tables_agent_read_UNVERIFIED.csv and not used “until the units are settled from the chapter's figures”. The units are settled by Figures 15 and 16 (X0 in µm, t/X0 in h/µm): B in µm²/h, B/A in µm/h. The transcription's values agree with the second reading's, entry for entry.
• Gupta and Akhtar 2011. The transcription keeps the printed “1050” for the fourth column; the text (p. 15, p. 19) and Figure 10 make it 1150 °C.
• Goto and Hijikata 2016. The project held this paper as “abstract read … nothing digitized”, with no stored recipe and no dataset. The rates and parameters here are the second reader's readings of Figures 3 to 5.
• Goto and Hijikata 2016. Text p. 6: activation energies of k0 “0.68, 1.37 and 2.02” for the “Si-, a- and C-faces”; Figure 4(a) labels 2.02 eV on the Si-face line and 0.68 eV on the C-face line, and the paper's own argument (three Si back-bonds on the Si face, one on the C face) requires the figure's order.
• Li, Luo and Ye 2023. There is no stored recipe or dataset for this paper (ontology/library.json: status skimmed, used by nothing). Everything here is from the second reading; nothing contradicts an earlier encoding because there is none.
• Li, Luo and Ye 2023. SOURCE.md says “carbon accumulation at the interface is reduced by the implantation” from the abstract. Figure 8 shows no carbon excess above the SiC level on either sample: the “accumulation” is a wider transition (3.6 against 2.7 nm, digitised), not a peak.
• Hijikata, Yagi, Yaguchi and Yoshida 2013. There is no stored recipe or dataset for this chapter (ontology/library.json: status skimmed, used by nothing). SOURCE.md lists Figures 3, 5, 7 and 9 as the ones with data; Figure 1 (C face at 8 temperatures, linear axes) is the most complete and is used here.
• Hijikata, Yagi, Yaguchi and Yoshida 2013. Figure 3's Si face at 1100 °C and Figure 5's 1100 °C curve agree point for point; they are taken to be the same run and only Figure 5 is used.
• Long et al. 2012. SOURCE.md and ontology/library.json describe the paper as “photoemission depth profiles at two temperatures (Figures 1 to 3)”. The profiles that carry the paper are RBS (Figures 5-7) at five temperatures; XPS (Figures 1-3) shows chemistry at two.
• Long et al. 2012. The inventory program's comment says it follows Long p. 6. Long's Eq. 7 counts the whole germanium in a box-shaped pile-up of uniform X_pileup; the program puts the same dose, x N_ox X/(1−X), as the excess above X in an exponential. The two count different germanium (see the reading).
• Thornton et al. 2022. SOURCE.md and ontology/library.json say “at 900 C”. The paper oxidizes at 750, 800, 850 and 900 °C and its result is an Arrhenius plot over the four.
• Thornton et al. 2022. The challenge brief and library.json call these SiGe fins; the germanium the paper measures moves down the silicon base of a Si/SiGe superlattice fin. The oxidizing interface where the length is measured is Si/SiO2, fed from SiGe layers above.
• Fukuda et al. 1992. The project held no stored recipe for this paper: scripts/analyse_fukuda.py fits closed-form laws to Figure 3 with the oxidation time as the only history. This page adds the heat-up, cooling, gate deposition and doping as RUNs, all with stated or searched ranges.
• Fukuda et al. 1992. The project's Figure 3 reading (data/primary_sources/fukuda_1992_jap_72_1906/figure3_agent_read_UNVERIFIED.csv) and the second reading agree within 5% on all 14 points (for example 5.0 against 5.1×10^11 at 1000 °C, 10 s); Figures 2 and 7 agree within 5%.
• Kar and Narasimhan 1987. The project held no recipe and no digitized value for this paper. SOURCE.md says pages 2 and 5 were read and that the paper reports only the peak density near the band edges, “not comparable with midgap quasi-static values”. Figure 3 (PDF p. 3) does reach 0.50 to 0.58 eV for the 93 and 125 Å n-type devices (about 3.1 and 2.7×10^11 cm⁻² eV⁻¹ at 0.55 eV), so two midgap values exist, by another technique (admittance, not quasi-static).
• Kar and Narasimhan 1987. SOURCE.md does not mention Figure 1: a thickness-against-time series at 900 °C, 79 to 211 Å over 20 to 140 min, on the same wafers. It is a second laboratory's thin dry oxides at the temperature the thin-oxide program holds out (900 °C).
• Tsai, Wu, Wu and Hu 1987. The project holds no recipe or dataset for this paper (ontology/library.json: status held, used_by empty; SOURCE.md: “Not read by us”). Everything here is from the second reading.
• Tsai, Wu, Wu and Hu 1987. SOURCE.md says the paper reports “trap density after nitridation at 900 to 1150 C”. It reports the trap density generated during electron injection after nitridation (Figure 1). The pre-stress trap density was measured (“first measured for the interface state density”) but is not printed; the curves of Figure 1 start below about 1×10^11 cm⁻² eV⁻¹.
• O'Sullivan et al. 2001. No dataset and no script uses this paper; SOURCE.md says the figures and Table I were not digitized. The midgap values here are new graph readings and Table I is transcribed here for the first time.
• O'Sullivan et al. 2001. ontology/library.json gives the source's conditions as “pull: rapid thermal, 10 to 100 C/s” and “metal: mercury probe, no gate”. That describes processes 3–5 only. Processes 1 and 2 are polysilicon-gate capacitors whose oxides were furnace grown and pulled slowly (3 °C/min to 600 °C, then 8 cm/min, for process 1); only their last step is rapid.
• Reed 1989. The project had no dataset, script or stored recipe from this paper (library status “skimmed”); SOURCE.md says only the first page was read. Everything here is new.
• Reed 1989. A 6 s or 12 s furnace anneal cannot reach 450 or 570 °C in the wafer; the review writes them as anneal times without saying how the wafer was heated. Ours holds the wafer at temperature for the stated time.
• Kar, Shanker and Chari 1985. The source note (data/primary_sources/kar_shanker_chari_1985_apl_47_1203/SOURCE.md) says “two thicknesses at 900 C”. The paper has six thicknesses in each of three sets (two n-type, one p-type), 79 to 227 Å.
• Kar, Shanker and Chari 1985. The library lists this source as held and used by nothing, quantity trap_density. Its numbers are peak densities near a band edge (Figure 2), the same kind of quantity the library notes as not comparable with midgap values for Kar and Narasimhan 1987; only Figure 1 gives values near midgap, and only by eye.
• Akinwande, Ho and Plummer 1981. No stored recipe or dataset holds this report: ontology/library.json lists the source as “held … Not yet read”. Its as-grown and argon-annealed charges are in the project as the 1983 report's redrawings (data/primary_sources/stanford_suprem_report_ada130105/figures_1_2_6_7_agent_read_UNVERIFIED.csv, Figures 1 and 2), read there from a lower-resolution scan.
• Akinwande, Ho and Plummer 1981. The two readings agree. As grown, the 1983 redrawing's (111) values are within 5% of ours at every temperature (e.g. 3.5 against 3.4 × 10^11 at 750 °C, 1.53 against 1.55 at 1200 °C); (100) within 10% except 1100 °C (4.5 against 4.65 × 10^10). Argon-annealed: within 5% for (111); (100) 5.0 against 5.05 at 900 °C, 2.8 against 3.5 × 10^10 at 1100 °C (the two symbols there are 1.5 and 5.5, so the mean depends on the reading).

Aligning the simulations with the papers

For each study the steps a paper leaves open were searched over the range the paper allows (see each page). Counted on the numbers each page's alignment chart shows: where a study has held-out samples these are only the held-out ones, a test; elsewhere the unknowns were chosen on the same numbers and the aligned count is an upper bound.

What the language gained from reproducing these studies

Each addition below was forced by a paper that could not otherwise be written down as it was done. The words are in ontology/isl_vocabulary.json (history_statements) and the language reference, and the runtime refuses anything else.

What our language and ontology still cannot say

Collected from every page. Each is something a paper states, or a simulation needs, that cannot yet be written down; each says what it would change. This is the agenda for the next changes to the language.

• Done 10 November: a furnace run shared by several samples. A history may now hold several SAMPLE blocks under one CONTROL_TRAJECTORY, and scripts/align_study.py can share an unknown per run (share = "run"). Still missing: a record in the ontology that measurements came from one run. Before: The result above depends on four wafers having one history. A history file has one SAMPLE; the ontology has no record that measurements came from one run. A BATCH of samples under one CONTROL_TRAJECTORY would let one unknown be declared once and inferred from all of them. ([Vitkavage, Irene and Massoud 1990](/writing/interface-study-vitkavage-1990))
• Done 10 November: an unknown with a range, as a statement. TIMING bound and VARY duration 0.5..5 now exist and drive the alignment search. Before: “Shorter than 5 sec” had to become DURATION 5. The language needs DURATION 0.5..5 (a bound), and a status on each step's timing (stated, bound, assumed), which on this page is hand-written in Python. Then a scan like this one is a query, not a script. ([Vitkavage, Irene and Massoud 1990](/writing/interface-study-vitkavage-1990))
• Partly done 10 November: cooling as a law. PROFILE exponential TAU now exists; the charge program still needs a rate that freezes as the wafer cools. Before: A RUN's temperature is linear from start to end. A wafer leaving a furnace cools toward the room with a time constant. Without that, the end temperature of a pull (400 °C) is a cut-off that stands in for physics, and the charge program, whose oxygen rate barely depends on temperature (0.05 eV), keeps moving charge at 100 °C if a cooling step is written. ([Vitkavage, Irene and Massoud 1990](/writing/interface-study-vitkavage-1990))
• Done 10 November: steps with no duration. EVENT <id> KIND <kind> now records them in place. Still missing: an ontology class a program could read. Before: Clean, load, storage, metallization and measurement are comments in the history. The ontology has no classes for them that a program could read; an EVENT statement with a kind from the vocabulary would at least make them queryable. ([Vitkavage, Irene and Massoud 1990](/writing/interface-study-vitkavage-1990))
• Sample properties a program can read. Doping, wafer size, starting oxide and sample set are recorded and read by nothing. The starting oxide in particular is a parameter inside the growth program (23 nm) when it is a property of the wafer. ([Vitkavage, Irene and Massoud 1990](/writing/interface-study-vitkavage-1990))
• Done 10 November: an orientation with no constants. It now runs on the (100) line and is labelled outside the tested scope. Before: (110) and (511) fell back to (111) for fixed charge with a note; the number looked like any other. ([Vitkavage, Irene and Massoud 1990](/writing/interface-study-vitkavage-1990))
• Partly done 10 November: gas by pressure, not by name. The predictor now reads a gas by its partial pressures. Still open: the thin-oxide program grows oxide in nitrogen because it does not read the oxygen pressure. Before: Writing the stated moisture (mix:O2=0.999994,H2O=6.0e-6) made the fixed-charge program return nothing, because the predictor looks for the name dry_O2. And the thin-oxide program keeps growing oxide during a nitrogen step: it does not read the oxygen pressure. ([Vitkavage, Irene and Massoud 1990](/writing/interface-study-vitkavage-1990))
• What a measurement is. Fixed charge after a hydrogen anneal, apparent charge without one, and Deal's 1967 charge after 550 °C in nitrogen are three observables in the papers and one in ours. The work-function value behind every fixed charge here is not stated; an observation model for the C-V extraction would carry that assumption. ([Vitkavage, Irene and Massoud 1990](/writing/interface-study-vitkavage-1990))
• Repeatability as data. The report's run-to-run scatter (up to 33%) is larger than the 10% we score against. A dataset should carry its own repeatability so a contract's tolerance comes from the source. ([Vitkavage, Irene and Massoud 1990](/writing/interface-study-vitkavage-1990))
• A pull as a motion through a furnace profile: PULL DISTANCE 54 cm DURATION t PROFILE <furnace axial temperature profile>, so that the wafer's temperature against time follows from the paper's own words (time and distance) instead of an end temperature and a shape we invent. The paper gives distance and time; we have to write a temperature ramp. ([Deal, Sklar, Grove and Snow 1967](/writing/interface-study-deal-1967))
• A history that ends at the measurement temperature by default (a final cooling to 25 °C written explicitly, with its gas). Our histories stop at the end of the last hot step, which is what hid the low-temperature behaviour of the charge program. ([Deal, Sklar, Grove and Snow 1967](/writing/interface-study-deal-1967))
• A way to search a PROFILE TAU as an unknown (VARY tau <lo>..<hi>, or VARY tau_factor relative to the duration). The exponential pull had to be tried as a handful of declared variants. ([Deal, Sklar, Grove and Snow 1967](/writing/interface-study-deal-1967))
• A tag for which samples were measured with a gold probe on bare oxide and which were metallized and given the 550 °C treatment: an optional step (EVENT … OPTIONAL, or two histories under one measured value), since the paper does not say which tabulated value is which. ([Deal, Sklar, Grove and Snow 1967](/writing/interface-study-deal-1967))
• An ambient for “the end of the tube”: where the slice sits after the pull (open tube, flowing oxygen, room air) is part of the history and has no word. ([Deal, Sklar, Grove and Snow 1967](/writing/interface-study-deal-1967))
• Cooling with a freezing point. With PROFILE exponential the language can now cool a wafer realistically, but the charge program then keeps moving charge down to room temperature, because its oxygen rate was fitted with an activation energy of 0.05 eV. A program needs a rate that vanishes as the wafer cools, fitted against pulls written as real cooling curves; then the pull's end temperature stops being a free number. ([Deal and Razouk 1980](/writing/interface-study-deal-razouk-1980))
• Doping as an input. DOPING is recorded on each sample and read by nothing. The report's n-type and p-type wafers differ by more than the tolerance. ([Deal and Razouk 1980](/writing/interface-study-deal-razouk-1980))
• The anneal's gas as a separate level. Nitrogen and argon are one curve in the report and one level in the program; if they differ (the 1100 °C rows hint that they might), the program needs a level per inert gas. ([Deal and Razouk 1980](/writing/interface-study-deal-razouk-1980))
• Which step is the oxidation. The fixed-charge and trap-density programs take the oxidation temperature from the first oxygen-bearing RUN. A boat-in written as a heating ramp (TEMP 25->1050 PROFILE exponential) would then be read as an oxidation at 25 °C, so the boat-in here had to be written as a hold at furnace temperature. A ROLE oxidation|anneal|load|pull on each RUN would let programs find the step by what it is, and let a push-in be written as the heating it is. ([Hung 1987](/writing/interface-study-hung-1987))
• An unknown that is an offset, not a duration. “Does the printed time include the 3 min boat-in?” is one question for all 24 wafers whose oxidation times run from 1.5 to 151 min. VARY duration on the boat-in stands in for it; VARY offset -180..0 on the oxidation, shared across samples, would say it directly. ([Hung 1987](/writing/interface-study-hung-1987))
• A choice of gas as an unknown. The ramp from 900 to 1050 °C and the withdrawal after the anneal series have unknown gases. The language varies numbers only; the gas had to become a Python variant. VARY ambient inert_N2|dry_O2 would make it a declared unknown. ([Hung 1987](/writing/interface-study-hung-1987))
• A wafer split in two. In chapter 5 each wafer was cleaved and only one half had forming gas; Figure 5.9's trap density is from one half and its fixed charge from the other. EVENT Cleave records it, but one history cannot carry two halves. A SPLIT that forks the trajectory into named halves, with each observation bound to its half, would let both values be compared on the right history. ([Hung 1987](/writing/interface-study-hung-1987))
• An etch that removes oxide. The etched-back oxides are Hung's control: same growth, different final thickness. EVENT EtchBack KIND etch takes nothing off the simulated oxide. A statement that sets or removes thickness (ETCH REMOVE 25 nm or ETCH TO 14.5 nm) is needed before any program with a thickness term can be tested against this control. ([Hung 1987](/writing/interface-study-hung-1987))
• How a charge was extracted. Hung's Qf uses φMS = -0.715 V from the etched-back intercept. At 10 nm, 0.1 V of φMS is 2.1 × 10^11 cm⁻² of charge; at 40 nm it is 0.5 × 10^11. An observation statement carrying the φMS, the frequency and whether the half had forming gas would let charge from different laboratories be compared on one basis. ([Hung 1987](/writing/interface-study-hung-1987))
• Gas by pressure in every program. Writing mix:O2=0.2,N2=0.8 now reaches the programs, but the thin-oxide growth program ignores the oxygen pressure and grows oxide in nitrogen. That is a predictor change, not a language one, and it decides every 20% and 50% thickness on this page. ([Hung 1987](/writing/interface-study-hung-1987))
• Rounded and typical values. Table 4.2 gives “~5×10^10” for a group of four, and chapter 5 reports “only typical or representative results”. A result's tolerance should be able to say it is a rounding of a group value, so that a search does not fit one wafer to it, as happened with the forming-gas time here. ([Hung 1987](/writing/interface-study-hung-1987))
• A clock reference for a RUN: TIME_ZERO at_load | at_temperature | at_gas_on (or a signed OFFSET <s>), so that “time” on a figure axis can be stated as what it is. Here the offset had to be an extra 1 s step, because a RUN cannot have zero duration and VARY cannot go negative. ([Song, Dhar, Feldman, Chung and Williams 2004](/writing/interface-study-song-2004))
• Measurements part-way through a history: MEASURE <observable> AFTER <step> followed by more steps, so a cumulatively oxidized sample (oxidize, unload, measure, reload, oxidize) is one history with several measured points. Today each point has to be a separate sample built by a variant function. ([Song, Dhar, Feldman, Chung and Williams 2004](/writing/interface-study-song-2004))
• An optional or repeated step (REPEAT <n> { … }) for the reload of a cumulative series, with its own VARY on the per-reload time. ([Song, Dhar, Feldman, Chung and Williams 2004](/writing/interface-study-song-2004))
• OFFCUT as a field a program reads, so that this on-axis data and Kobayashi's 4° off-axis epilayers are not the same sample to the growth program; and a word for the measurement technique (RBS or ellipsometry) on an observable. ([Song, Dhar, Feldman, Chung and Williams 2004](/writing/interface-study-song-2004))
• A cooling rate as a statement (COOL RATE <K/min> TO <T> [BOUND upper|lower]), so “< 70 °C/min” is written as the paper gives it rather than as an invented end temperature and duration. The rate also needs its own furnace, so a cooling difference is not silently also a furnace difference. ([Kobayashi et al. 2017](/writing/interface-study-kobayashi-2017))
• A furnace or equipment field on a RUN (FURNACE resistive | rf_induction), and the residual oxygen it is known for (under 1 ppm for one, not reported for the other). Here the two cooling rates are two furnaces. ([Kobayashi et al. 2017](/writing/interface-study-kobayashi-2017))
• A transfer through air as a RUN-like step with its own gas and time (TRANSFER AMBIENT air DURATION …): the Transfer here is an EVENT, so any exposure of the interface to room air between furnaces cannot reach a program. ([Kobayashi et al. 2017](/writing/interface-study-kobayashi-2017))
• OFFCUT read by programs (4° off-axis here, on-axis in Song 2004), and the measurement method on an observable (1 MHz high-low counts fewer states than a 100 MHz reference, the paper's Figure 4). ([Kobayashi et al. 2017](/writing/interface-study-kobayashi-2017))
• For the predictor: a small tolerance on a state's declared bounds, or a hard clamp, so that a long low-temperature hold does not refuse a history at exactly 1.00. ([Kobayashi et al. 2017](/writing/interface-study-kobayashi-2017))
• An isotope in a gas: mix:18O2=0.98,16O2=0.02 PRESSURE 0.01 Torr, so that the enrichment and the total pressure are stated and a program can read them. Today the isotope is implicit in the program and the pressure, though written, reaches nothing. ([Goncharova et al. 2011](/writing/interface-study-goncharova-2011))
• A layer in the sample with its own composition that a program reads as an initial state (LAYER interlayer SiOxNy THICKNESS 6 Å O:N 5:1), so that the 6 Å oxynitride and its nitrogen start in the interface reservoir instead of being left out. ([Goncharova et al. 2011](/writing/interface-study-goncharova-2011))
• A check in the predictor that a program's operators can act in the history's gas: a program whose source term needs oxygen should refuse, or warn, on a RUN in vacuum or argon instead of acting as if the gas were present. ([Goncharova et al. 2011](/writing/interface-study-goncharova-2011))
• Measurement in the middle of a history (MEASURE … AFTER Anneal10min, then more steps): Figure 2's 10 min and +30 min are one film measured twice, which today must be two separate samples. ([Goncharova et al. 2011](/writing/interface-study-goncharova-2011))
• Temperature in K as a unit on TEMP (TEMP 763 K): 763 K is written 489.85 °C, and the paper's own 763 K, 773 K and 490 °C for the same anneal are easier to keep apart when the unit is the paper's. ([Goncharova et al. 2011](/writing/interface-study-goncharova-2011))
• A rate that freezes. PROFILE exponential now lets a history say how a wafer cools. This study shows the language cannot yet stop a program from moving charge at room temperature. The ontology could carry, per operator, a temperature (or activation energy) below which it is declared inactive, and the predictor could refuse a history that runs an operator outside it. That would turn the hidden 400 °C cut-off into a stated, testable property. ([Akinwande, Ho and Plummer 1982](/writing/interface-study-stanford-1982))
• The gas a wafer cools in. A pull leaves one gas and cools in another (oxygen in the tube, nitrogen at the flushed mouth, air in the room). A RUN has one AMBIENT. A PROFILE that names the gas after withdrawal, or a WITHDRAW step with its own gas, would make the variant that scores best here a statement. ([Akinwande, Ho and Plummer 1982](/writing/interface-study-stanford-1982))
• HCl in a mixture. “20 min 1% HCl/Ar etch” cannot be written: the mixture gases are O2, N2, H2, H2O and Ar. Adding HCl (and Cl2) would record the etch and, later, the HCl oxidations of Razouk and Deal's Figure 9. ([Akinwande, Ho and Plummer 1982](/writing/interface-study-stanford-1982))
• One wafer measured several times. The cycling is one sequence of anneals with a measurement after each. Five separate histories hide that they are prefixes of one. A MEASURE event inside a history whose readings are the scored values would state it once and would show that the metal must come off between measurements. ([Akinwande, Ho and Plummer 1982](/writing/interface-study-stanford-1982))
• Where a starting value comes from. The charge program's initial state is Deal's 1967 line for the oxidation temperature. Stanford's as-grown values sit below it on (111). The starting point is a laboratory property hidden in a program, and should be a stated input (for example INITIAL Q_f FROM laboratory with the 1967 line as one choice). ([Akinwande, Ho and Plummer 1982](/writing/interface-study-stanford-1982))
• Time zero of an anneal. Figure 5 times count from an unstated moment (the push, or reaching temperature). TIMING says whether a duration is stated, not what it is counted from. An ORIGIN push|at_temperature|gas_switch on a RUN would carry it. ([Akinwande, Ho and Plummer 1982](/writing/interface-study-stanford-1982))
• Repeat durations. The second 25 h and 20 min of the cycling are assumed equal to the first. A TIMING assumed_same_as <step> would make the assumption visible. ([Akinwande, Ho and Plummer 1982](/writing/interface-study-stanford-1982))
• A laboratory as a property of the sample. The largest single error here (4 to 7 times) is the laboratory. The ontology has the process factor laboratory but no sample or history field that records where a wafer was made and measured, so no program can carry a laboratory level, and the dataset cannot say which numbers share one. ([Razouk and Deal 1979](/writing/interface-study-razouk-deal-1979))
• An unknown that is a measured quantity. The oxidation time here should be inferred from a thickness the paper states. VARY lets us search the time, but the range had to be computed by hand from the 1965 law; a statement INFER duration FROM Thickness 120..220 nm would let the language derive it and record that it was derived. ([Razouk and Deal 1979](/writing/interface-study-razouk-deal-1979))
• Halves of one wafer. Every value here comes in pairs: one half annealed in hydrogen, one not. Two separate SAMPLEs lose that they share an oxidation; a CLEAVE that forks one history into two would carry it. ([Razouk and Deal 1979](/writing/interface-study-razouk-deal-1979))
• Cooling in an end-piece. “Pulled … in less than 3 sec into a quartz elephant” is a fast withdrawal followed by an unknown dwell in a closed, still oxygen-filled quartz piece. A RUN with PROFILE exponential can hold the cooling but the language has no word for where the wafer is. ([Razouk and Deal 1979](/writing/interface-study-razouk-deal-1979))
• Mixtures with HCl and pyrogenic steam. Figure 9's 5% and 10% HCl in oxygen cannot be written (the mixture gases are O2, N2, H2, H2O, Ar), so that figure is left off this page. ([Razouk and Deal 1979](/writing/interface-study-razouk-deal-1979))
• Doping read by a program. Figure 8's n- and p-type (100) differ by 40% at 1100 and 1200 °C; DOPING is recorded on every sample and read by nothing. ([Razouk and Deal 1979](/writing/interface-study-razouk-deal-1979))
• The predictor: the trap program does not respond to the gas of a step or to a pull, and labels (111) without hydrogen “outside the tested scope” while still reporting a number 35 to 48 times low. A refusal would be more honest than a number. ([Razouk and Deal 1979](/writing/interface-study-razouk-deal-1979))
• A starting oxide on the sample. The 230 Å is a PARAMETER of the program and applies to every history, wet included, where the paper's own start is zero. A sample field such as STARTING_OXIDE: 0 nm (read by the growth programs, with the program's intercept as its default) would let wet and dry slices of one preparation differ, and would let the alignment search it. ([Deal and Grove 1965](/writing/interface-study-deal-grove-1965))
• Water as an oxidant. wet_O2 has no partial pressures; written as a mixture the water reaches no program, and the 1965 program grows the 16% oxygen share. A wet growth program (Table I constants, B with 16.3 kcal/mole) reading p_H2O is needed before any wet history means anything. ([Deal and Grove 1965](/writing/interface-study-deal-grove-1965))
• Pressure on B only. The paper's law has B proportional to the oxidant pressure and A independent of it. Our operator scales the whole rate. A program that reads p_O2 only in B would be the paper's law for mixtures, which Study B (Krzeminski's 10% oxygen ramp) needs. ([Deal and Grove 1965](/writing/interface-study-deal-grove-1965))
• One slice measured several times. The language cannot say that a curve's points are the same slice taken out, measured and put back (an EVENT unload, measure, reload between RUNs of one sample) or separate slices. The paper does not say which; the language should at least be able to write both. ([Deal and Grove 1965](/writing/interface-study-deal-grove-1965))
• A support label per gas. The planar growth program's label claims wet and steam steps grow nothing, but a mixture makes it grow the oxygen share. The label should be computed from what the program reads (p_O2) rather than from the gas class. ([Deal and Grove 1965](/writing/interface-study-deal-grove-1965))
• A starting oxide a program reads. The single most useful addition. The thin-oxide program starts every history at its fitted 1.71 nm, so Massoud's 1.0 nm wafers, Krzeminski's HF-dipped wafers and a 1.53 nm post-ramp oxide all start the same. With STARTING_OXIDE read by the growth programs (default: the program's fitted value), the search's other-laboratory test (start at the measured 1.53 nm) and the predictor would give one answer instead of two. ([Krzeminski et al. 2007](/writing/interface-study-krzeminski-2007))
• Growth programs that read the oxygen pressure. The thin-oxide program has no p_O2 in its rate. It grows in nitrogen (0.2 nm in a 30 min reflow at 675 °C) and grows a 9% oxygen ramp as pure oxygen. Multiplying B (or the power law's constant) by p_O2/p_ref, as the 1965 program does, would make the stated cycle meaningful; the alignment's ramp of 580 s is a symptom. ([Krzeminski et al. 2007](/writing/interface-study-krzeminski-2007))
• Time zero as a statement. The datasets count time from the start of the main stage; the language counts from the first RUN. A marker such as EVENT TimeZero KIND other that results can refer to, or a t0 on the dataset rows, would let the stored recipe hold the whole cycle without moving every time. ([Krzeminski et al. 2007](/writing/interface-study-krzeminski-2007))
• A curve of separate wafers. Each of the paper's points is a wafer that went through the whole cycle; Massoud's are one wafer watched in situ. The study format can sample in situ (curves within a RUN) but needs one case per wafer for the ex-situ points. The language should say which kind of series a dataset is. ([Krzeminski et al. 2007](/writing/interface-study-krzeminski-2007))
• A fast first regime. Neither the 1965 law nor the searched power law grows the first 1.5 nm of a bare surface; both fold it into a starting thickness. An operator for it (or a second state for the first monolayers) is what Krzeminski's ramp asks for, and what its own reaction-rate model provides. ([Krzeminski et al. 2007](/writing/interface-study-krzeminski-2007))
• Exposure to air before an ex-situ measurement. An uncaptioned figure shows 0.25–0.37 nm of growth in 90 min of air. A storage EVENT with a duration and an ambient (room air) that a program can read would carry it. ([Krzeminski et al. 2007](/writing/interface-study-krzeminski-2007))
• Growth in dry oxygen and temperature in the nanowire program. The paper's cycle puts the wires through 750 → 850 °C and 5 min at 850 °C in pure oxygen before any steam. The radial program's rate is p_H2O × constant, so these steps grow nothing, and the missing oxide reappears as a wet time 77% longer than stated. A dry channel (rate ∝ p_O2 with its own constant) and an Arrhenius temperature dependence would let the stated cycle be simulated as stated, and would let the ramp unknowns be inferred instead of being invisible. ([Müller, Lecestre, Demoulin, Cristiano, Hartmann and Larrieu 2023](/writing/interface-study-muller-2023))
• A rate constant that carries the gas it was measured in. The program's rate was measured in this furnace's pyrogenic steam, then multiplied by p_H2O/1 atm. Writing the gas truthfully as a mixture lowers the rate. A PARAMETER should be able to say “measured at p_H2O = the furnace's own” so that a stated mixture and the calibration are not counted twice; today the steam fraction and the wet time are interchangeable unknowns. ([Müller, Lecestre, Demoulin, Cristiano, Hartmann and Larrieu 2023](/writing/interface-study-muller-2023))
• A starting oxide on the sample. Figure A1's bulk line extrapolates to about 9 nm at zero wet time; the wires had an HF strip then a 20 W oxygen plasma. Neither program can start a wire with an oxide shell. A sample field OXIDE_NM, read as the initial shell, would separate the oxide grown before time zero from the oxide grown in the stated time. ([Müller, Lecestre, Demoulin, Cristiano, Hartmann and Larrieu 2023](/writing/interface-study-muller-2023))
• Shape must select the program. The predictor does not read SHAPE: a nanosheet stored with DIAMETER_NM is run as a cylinder, and a SiGe nanosheet is run by planar programs (labelled outside scope, but still numbers). SHAPE: nanosheet should run a flat-sidewall program (the radial program with a → ∞ gives one) or refuse. ([Müller, Lecestre, Demoulin, Cristiano, Hartmann and Larrieu 2023](/writing/interface-study-muller-2023))
• Gas in the SiGe growth program. The Rabie program grows at its wet-oxidation rate in any gas, including the dry-oxygen ramp and the nitrogen steps' temperature. It should refuse, or read p_H2O and p_O2, before the furnace-cycle unknowns can be searched on SiGe. ([Müller, Lecestre, Demoulin, Cristiano, Hartmann and Larrieu 2023](/writing/interface-study-muller-2023))
• Furnace runs as data. All Si structures went through one run and both SiGe compositions through another, at different times chosen per material. The alignment shares unknowns per run (case["run"]), but the history language has no statement for it; a BATCH of samples under one trajectory would put this in the history itself. ([Müller, Lecestre, Demoulin, Cristiano, Hartmann and Larrieu 2023](/writing/interface-study-muller-2023))
• An observable for what the paper measures. Figure 5(a) is the increase of the outer diameter; Figure 5(b) is the core after a buffered-oxide etch. The dataset turns them into an oxide thickness, which already assumes the etch removes no silicon. Observables Outer_diameter and Core_diameter_after_strip would let the program be scored on the measured quantities and show the volume balance the data do not close. ([Müller, Lecestre, Demoulin, Cristiano, Hartmann and Larrieu 2023](/writing/interface-study-muller-2023))
• Faceting and sidewall orientation. Above 60 nm the paper's Si wires are faceted by faster growth on {110}. A wire has one rate around its circumference in our language; a sidewall orientation mix on the sample would let that be tested. ([Müller, Lecestre, Demoulin, Cristiano, Hartmann and Larrieu 2023](/writing/interface-study-muller-2023))
• A reference sample in the same run. The nanobeams were oxidized beside the wires; their oxide is the furnace's flat-surface rate. The language cannot say that two samples shared a furnace run, nor let a program take its rate from a measured co-processed sample. Here it had to be smuggled in as a steam fraction per case. A BATCH of samples under one trajectory, with a REFERENCE sample whose measured value is an input, would make that explicit and stop the steam fraction meaning two things. ([Krzeminski, Han and Larrieu 2012](/writing/interface-study-krzeminski-2012))
• A starting oxide on the sample. The ~5 nm at t = 0 has no place: the radial program starts from bare silicon, and writing it as extra steam time grows it radially with the wire's slowdown. A sample field OXIDE_NM read as the initial shell (b > a at t = 0) would let the 5 nm be tested as native or ramp oxide. ([Krzeminski, Han and Larrieu 2012](/writing/interface-study-krzeminski-2012))
• Shapes beyond a cylinder. Nanobeams (flat sidewalls) and nanorings (concave) are half of this paper's data and have no SHAPE a program reads. A SHAPE: nanobeam should select a flat-sidewall program, and a SHAPE: nanoring with an inner radius should select the concave sign of Eq. 1; today a beam stored with DIAMETER_NM is silently run as a cylinder. ([Krzeminski, Han and Larrieu 2012](/writing/interface-study-krzeminski-2012))
• Temperature and gas in the nanowire program. Its rate depends on p_H2O only, so ramps, dry oxygen and the 850 °C itself are invisible. An Arrhenius rate and a dry-oxygen channel would let the furnace-cycle unknowns mean something. ([Krzeminski, Han and Larrieu 2012](/writing/interface-study-krzeminski-2012))
• Which form runs. The predictor runs the viscous shell fitted on one paper while the project's own search selected another form on two. A program registry that records which form a search selected, and runs that one, would make the page and the search agree. ([Krzeminski, Han and Larrieu 2012](/writing/interface-study-krzeminski-2012))
• Wet oxygen by its steam pressure, or by its source. wet_O2 carries no partial pressures. A history should be able to say mix:H2O=x,O2=1-x, as here, or name the source (BUBBLER 95 C, PYROGENIC H2:O2 1.8) from which the pressure follows. Here the fraction moves the layer extent from 47 to 92 nm. ([Rabie, Haddara and Carette 2005](/writing/interface-study-rabie-2005))
• Constants that belong to a condition, and a refusal outside it. The program's k1, k2, k3 and DSiGe are the paper's for 800 °C, wet oxygen, 14%. A parameter should carry the conditions it was chosen for (VALID T 800; AMBIENT wet; GE 0.14), so the predictor labels or refuses a 700 °C steam history instead of answering it. ([Rabie, Haddara and Carette 2005](/writing/interface-study-rabie-2005))
• Programs that read the gas. The Rabie program ignores the ambient entirely: a push or a cooling in nitrogen would grow oxide in it, and dry oxidations (David 2015) are run with wet-oxygen constants. ([Rabie, Haddara and Carette 2005](/writing/interface-study-rabie-2005))
• A starting oxide on the sample. The model's initial oxide (“very small thickness Lox”) is a 1 nm hypothesis inside the program. The early LeGoues points ask for about 10 nm; a STARTING_OXIDE field would make that an unknown the alignment can search. ([Rabie, Haddara and Carette 2005](/writing/interface-study-rabie-2005))
• Report every observable a program computes. The Rabie program computes the germanium at the front (51.5% here) and the predictor drops it, reporting the enrichment relation's 44% under the same name. Two programs answering one target should both be reported, each with its name. ([Rabie, Haddara and Carette 2005](/writing/interface-study-rabie-2005))
• A measurement with no time. “A 50-nm-thick Ge-rich layer was obtained with up to 75% Ge” has no oxidation time. A result should be able to say AT unknown (or a range of times) and be compared over that range, not at the time we chose. ([Rabie, Haddara and Carette 2005](/writing/interface-study-rabie-2005))
• An upper value as data. “Up to 75%” is scored as 75 ± 10%. A result should carry ≤, ≥ or about, so that a program giving 60% is not counted as wrong by the same rule as one giving 44%. ([Rabie, Haddara and Carette 2005](/writing/interface-study-rabie-2005))
• A finite alloy layer. The source experiments are epilayers of unstated thickness; Table VI consumes up to 425 nm of alloy. The sample cannot say how thick the alloy is or what lies under it. ([Rabie, Haddara and Carette 2005](/writing/interface-study-rabie-2005))
• A layered sample. “34 nm Si0.8Ge0.2 on 12 nm Si on 10 nm BOX” is recorded as a FILM string that nothing reads. The sample needs layers with thickness and composition (LAYER SiGe { COMPOSITION: Si0.8Ge0.2; THICKNESS: 34 nm; STRAIN: compressive }) and programs that stop consuming a layer when it is gone. This paper's whole second half (6 h, 13 h, 14 h, 15 h) depends on it. ([David et al. 2015](/writing/interface-study-david-2015))
• Constants tied to a gas and a condition, with a refusal outside them. The predictor answers a dry 950 °C oxidation with wet 800 °C constants and labels it “direction only”. Parameters should carry the conditions they were chosen for, and a history outside them should be labelled or refused. ([David et al. 2015](/writing/interface-study-david-2015))
• One oxide per history. Two programs report on one SiGe history and grow oxide 16 times apart (Rabie 77.5 nm, inventory about 5 nm at 800 °C, 90 min). The predictor should take the oxide from one program and feed it to the others, or report the disagreement. ([David et al. 2015](/writing/interface-study-david-2015))
• Scope read from the oxidation, not from every step. A ramp from room temperature puts every number “outside the tested scope”. Scope should be judged on the steps that do the work (the hold), with ramps reported separately. ([David et al. 2015](/writing/interface-study-david-2015))
• A furnace cycle as a unit. “By steps of 30 min in accordance with the RTO technical specifications” is a repeated cycle (ramp, 30 min, ramp). A REPEAT n { … } block, or a tool with its own ramp law, would write it in one line instead of 90 RUNs. ([David et al. 2015](/writing/interface-study-david-2015))
• VARY on a time constant. A cooling written as PROFILE exponential TAU cannot have its TAU searched; here the cooling had to be written as a straight line to make it an unknown. ([David et al. 2015](/writing/interface-study-david-2015))
• Germanium lost as GeO2. The paper reports under 10% lost by oxidation (about 26% from its own thicknesses); the inventory conserves germanium exactly. A loss flux, or at least a declared bound on it, belongs in the inventory program. ([David et al. 2015](/writing/interface-study-david-2015))
• Strain as a sample property. Relaxed against fully strained is the paper's subject (Eqs. 1 and 2); the language has no field for it that a program could read. ([David et al. 2015](/writing/interface-study-david-2015))
• A starting oxide on the sample. The paper's two figures disagree on it (0 or 1 nm) and our program holds it as a fitted parameter. A SAMPLE … STARTING_OXIDE <nm> field that growth programs read would make it an unknown the alignment can search and a fact a paper can state. ([Wang, Hu and Irene 1994](/writing/interface-study-wang-1994))
• Growth programs that read the oxygen pressure. The germanium program grows oxide at full rate in vacuum and at half an atmosphere. Every growth law should multiply by a pressure dependence (linear in p_O2 for the parabolic term at least); then the heat-up, the fill and the cooling in vacuum become testable instead of inert. ([Wang, Hu and Irene 1994](/writing/interface-study-wang-1994))
• A pressure that changes within a step. “Oxygen was admitted” into a closed chamber at 150 sccm is a ramp of pressure, as a ramp of temperature is. RUN … PRESSURE 0->1 (or a pressure PROFILE) would let the fill be written as stated, with the chamber volume as the unknown. ([Wang, Hu and Irene 1994](/writing/interface-study-wang-1994))
• A measured clock origin. The paper's time zero is the admission of oxygen, which is the start of the fill, not of our Oxidation step. A curve should name the event its clock starts at (CLOCK FROM Fill), so a fill or ramp in front does not shift every point. ([Wang, Hu and Irene 1994](/writing/interface-study-wang-1994))
• Two observables for one oxide. Total thickness of a two-layer optical model, the GeO2 layer alone, the interface layer (GeO2 + amorphous Ge, 1.4 to 4.8 nm), and a single-layer real-time value are four numbers for one sample here. The ontology has one Thickness. An interface-layer observable and an observation model per technique would let Table 1 be scored. ([Wang, Hu and Irene 1994](/writing/interface-study-wang-1994))
• Scope from the oxidation, not from the first oxygen step. The predictor labels a history by the start temperature of its first oxygen step, so a ramp in oxygen makes a 550 °C oxidation “outside the tested scope”. ([Wang, Hu and Irene 1994](/writing/interface-study-wang-1994))
• A germanium clean. The vocabulary of pre-oxidation cleans is silicon's; bromine-methanol with buffered HF cannot be named. ([Wang, Hu and Irene 1994](/writing/interface-study-wang-1994))
• The program as a file. The selected germanium law is generated in memory by scripts/search_growth.py; writing it to experiments/ as the SiC ones are would let this page print the model it ran. ([Wang, Hu and Irene 1994](/writing/interface-study-wang-1994))
• A reference sample whose measured value is an input, read as a thickness. RUN_FLAT_OXIDE_NM says “the oxide on a flat sidewall in the same run”, but the radial program reads it as oxide ÷ steam time and then applies its own diffusion term, so a 500 nm flat oxide becomes 349 nm on the program's flat sidewall. The field should mean what it says: the program should choose its flat rate so that its own flat sidewall grows the stated oxide (for this program, F = X(1 + X/A)). This one change moves the high-temperature cylinders from 0 of 5 to 2 of 5. ([Kao, McVittie, Nix and Saraswat 1988](/writing/interface-study-kao-1988))
• Temperature in the curved-silicon program. Kao's five temperatures are the point of the paper. A stress_time = 2η(T)ΔΩ/kT with an Arrhenius viscosity, and an A that follows Deal and Grove's temperature law, would let the program be scored at 800 to 1200 °C; today every temperature is 850 °C to it. ([Kao, McVittie, Nix and Saraswat 1988](/writing/interface-study-kao-1988))
• SHAPE: hole (concave). The language can write SHAPE hole; no program reads it. The radial program with the concave sign (Kao's Eqs. 8, 22, 37–40) would cover holes and trenches, which are half of Kao's data and all of Krzeminski's rings. ([Kao, McVittie, Nix and Saraswat 1988](/writing/interface-study-kao-1988))
• The radius after oxidation as a sample field and as an observable. Kao, like most curved-surface papers, reports the radius after oxidation. A field RADIUS_AFTER_NM (or an observable Core_radius scored against it) would let the program start from an unknown etched radius and be scored on the measured final one, instead of back-computing the start from the measured oxide. ([Kao, McVittie, Nix and Saraswat 1988](/writing/interface-study-kao-1988))
• A normalized observable. Kao reports curved oxide over flat oxide of the same run. An observable defined as a ratio of two samples' thicknesses in one BATCH would score exactly what was measured, without the (100)/(110) choice of the flat reference. ([Kao, McVittie, Nix and Saraswat 1988](/writing/interface-study-kao-1988))
• Sidewall orientation. Kao reads the (110) part of a wall whose orientation varies around the circumference; flat (110) grows up to 1.29 times flat (100) in steam. A field for the measured orientation that a curved program reads (the planar program already has r_or for 110) would remove one of the paper's ambiguities from the comparison. ([Kao, McVittie, Nix and Saraswat 1988](/writing/interface-study-kao-1988))
• A dry channel in the curved-silicon program. Kao's dry cylinders (Figure 10) cannot be predicted at all: the radial program's rate is proportional to the water pressure. The planar dry programs exist; the cylindrical law needs a dry B and B/A with an oxygen-pressure term, and RUN_FLAT_OXIDE_NM should give the dry rate as it gives the wet one. ([Kao, McVittie, Nix and Saraswat 1987 (Part I, report)](/writing/interface-study-kao-1987-report))
• The gas of a step as an unknown. Push and pull in nitrogen are invisible to every program; in steam or oxygen they would not be. The language can VARY a duration or a temperature but not the ambient; VARY ambient inert_N2|dry_O2|steam would let an unstated push gas be searched. ([Kao, McVittie, Nix and Saraswat 1987 (Part I, report)](/writing/interface-study-kao-1987-report))
• A steam source on the ambient. “Steam” in Kao, Deal and Grove and Müller is three different water pressures (bubbler 640 Torr, pyrogenic with leftover oxygen, unstated). An ambient word like steam_bubbler(95C) or steam_pyrogenic(H2:O2) that resolves to a partial pressure would make the choice explicit; here it moves the planar score from 3 to 5 of 18. And a program whose rate comes from RUN_FLAT_OXIDE_NM should not also scale by p_H2O, or a stated mixture counts the water twice (the bubbler variant drops the doped cylinders by 12%). ([Kao, McVittie, Nix and Saraswat 1987 (Part I, report)](/writing/interface-study-kao-1987-report))
• Doping and a glass on the surface as program inputs. DOPING and PSG_NM are recorded on the sample and read by nothing. Kao's doped groups differ by a factor of two in retardation at the same flat oxide; a stress-form program with a viscosity that depends on the phosphorus in the oxide, and a starting layer of a named material (not only STARTING_OXIDE_NM for SiO2), would let them be predicted. ([Kao, McVittie, Nix and Saraswat 1987 (Part I, report)](/writing/interface-study-kao-1987-report))
• A trench sidewall as a sample. The (110) values come from trench sidewalls; the language writes them as a flat (110) wafer. A SHAPE trench with a sidewall orientation would say what was measured. ([Kao, McVittie, Nix and Saraswat 1987 (Part I, report)](/writing/interface-study-kao-1987-report))
• An XPS observation statement. The paper measures Si(2p) shifts, intensity ratios and linewidths, not a profile. An OBSERVE XPS LINE Si2p TAKEOFF <deg> LAMBDA_OXIDE 2.5 nm LAMBDA_SI 2.3 nm that turns a simulated profile into δ and R would let a model be scored on the measured Figures 5 to 7 rather than on the authors' 0.25 nm-layer fit. ([Raider and Flitsch 1978](/writing/interface-study-raider-flitsch-1978))
• Liquid and chemical-oxidant ambients. Room-temperature water, boiling water, concentrated HNO3 and 30% H2O2 are half of the paper's oxidants. AMBIENT liquid:H2O or liquid:HNO3 (with temperature) would let them be written; today boiling water is approximated as steam and the acids cannot be written at all. ([Raider and Flitsch 1978](/writing/interface-study-raider-flitsch-1978))
• An etch that removes a layer. The 200 nm sacrificial oxide is grown and stripped in HF before every film. It had to be an EVENT because the language cannot take the oxide off again; a STRIP oxide or ETCH TO 0 nm would let the whole history be simulated, including what the first oxidation does to the silicon surface. ([Raider and Flitsch 1978](/writing/interface-study-raider-flitsch-1978))
• A group of samples with one pooled result. The width is fitted to all (100) films together and stated to be the same for each. A result attached to a group of histories (RESULT width 0.75 nm OVER SAMPLES 100_*), rather than repeated on each, would say what was measured. ([Raider and Flitsch 1978](/writing/interface-study-raider-flitsch-1978))
• Ranges of temperature as an unknown of the sample group. “100 to 700 °C” is a spread of samples, not an uncertainty about one; VARY T_hold 100..700 searches it as if it were one unknown value. A way to say “one history per value in this range” would match what was done. ([Raider and Flitsch 1978](/writing/interface-study-raider-flitsch-1978))
• Adsorbed contamination. A carbon layer of 0.2-0.6 nm on every surface (PDF p. 2) changes the XPS intensities; a sample field for an adsorbate layer would let an observation model account for it. ([Raider and Flitsch 1978](/writing/interface-study-raider-flitsch-1978))
• An ambient of atomic hydrogen. “Annealing … in atomic deuterium at 230 °C” downstream of a plasma has no word. It needs a species (H or D atoms) with a flux or an arrival rate at the surface (cm⁻² s⁻¹) rather than a partial pressure, since the atoms recombine on the walls; e.g. AMBIENT atomic:D FLUX 1e15. With it, a kinetic program could cap Pb at the rate atoms arrive at the interface, which this paper shows happens at 25 to 230 °C (Fig. 5). ([Johnson, Biegelsen and Moyer 1981](/writing/interface-study-johnson-1981))
• A film as a hydrogen source. The aluminium releases hydrogen during the anneal; the same anneal without it does nothing below 400 °C. A FILM Al { RELEASES: H } on the sample, read by hydrogen programs as a source term during any RUN, would let a post-metallization anneal in vacuum or nitrogen be written as what it is. The ontology has process factor hydrogen-anneal for a gas, nothing for a film. ([Johnson, Biegelsen and Moyer 1981](/writing/interface-study-johnson-1981))
• Isotopes. Deuterium is written as hydrogen or not at all. D2 in the gas list (and D in the atomic ambient) with masses carried into the zero-point energy (a computed ph.x number with deuterium masses) and into any rate prefactor would let the paper's “not regained up to 600 °C” be compared with Si–H dissociation rates on the right basis. ([Johnson, Biegelsen and Moyer 1981](/writing/interface-study-johnson-1981))
• A kinetic hydrogen program (see also Stesmans 1996): the trap program now has passivation from H₂ with Brower's 1.66 eV and release with Brower's 2.56 eV, sourced, not computed. This paper's with-Al series needs a passivation by atomic hydrogen besides, and its deuterated series a release slower than Brower's. Both barriers can be computed on the H100 with neb.x, from the relaxed cluster for H₂ + Pb and from a (111)/SiO₂ slab for dissociation into a bound site; the Stesmans page gives the plan. ([Johnson, Biegelsen and Moyer 1981](/writing/interface-study-johnson-1981))
• Spin density as an observable, with its units. Fig. 2 is in arbitrary units with one absolute anchor (6×10¹¹ cm⁻² as oxidized). A target Pb_spin_density in cm⁻² with a declared normalization (RELATIVE TO <case>) would let both programs be scored on Pb directly instead of on a share. ([Johnson, Biegelsen and Moyer 1981](/writing/interface-study-johnson-1981))
• A series on one sample. Whether the isochronal anneals of Fig. 2 are one specimen annealed in turn is not stated. The language cannot write either reading as one statement; REPEAT over temperatures with an observation after each pass would, and an alignment could then treat the reading as a variant. ([Johnson, Biegelsen and Moyer 1981](/writing/interface-study-johnson-1981))
• Interface states at an energy without a hydrogen step. The trap program reports a spectrum only after hydrogen, from the shape of one annealed curve. The as-oxidized (111) peak at 0.3 eV, 6.7×10¹² cm⁻² eV⁻¹, here and in Vitkavage, has no program at all. ([Johnson, Biegelsen and Moyer 1981](/writing/interface-study-johnson-1981))
• A kinetic hydrogen program: what it still needs. The trap-density program now has both forward rates with Brower's numbers: passivation Pb + H₂ → HPb + H (its fitted 400 °C rate times exp(−(Ef/k)(1/T − 1/T_H)), Ef 1.66 eV) and release HPb → Pb + H (kd = kd0·exp(−Ed/kT), Ed 2.56 eV, kd0 1.2×10¹² s⁻¹). Still missing: a state for the released hydrogen (in the oxide, at the interface, or recombined to H₂) so that the reverse rates follow from detailed balance with the computed reaction energies instead of a placeholder; a Gaussian spread of Ed (σ 0.105 eV), which needs a STATE distributed over a parameter (e.g. STATE N_HPb { DISTRIBUTED: Ed GAUSSIAN(2.63, 0.105) }) since one exponential cannot give the measured breadth with the measured barrier; and a creation term above about 650 °C in oxygen-free gas. With the barriers written PARAMETER Ed { STATUS: computed; SOURCE: evaluations/atomistic/<neb record> } instead of sourced from Brower, the measured values from Brower and this paper would be the test, not the input. ([Stesmans and Afanas'ev 1996](/writing/interface-study-stesmans-1996))
• Which barriers Quantum ESPRESSO can compute on the H100, and how. The cluster already relaxed (Si₄H₉/Si₄H₁₀, PBEsol, 40/320 Ry, spin-polarized) is the starting point; neb.x (climbing image, 7 to 9 images) does the rest. (1) Passivation, Pb + H₂ → HPb + H, on the cluster: H₂ approaching the dangling bond, final state HPb with a free H; this is small (15 to 16 atoms) and runs in hours, and gives a first number to set beside 1.66 eV. (2) Dissociation into a bound site: on the bare cluster HPb → Pb + H has no barrier beyond its 3.54 eV endothermicity, so the hydrogen needs somewhere to go. That needs an interface model: a (111) Si slab with a thin oxide (for example 4 to 6 Si bilayers under 1 to 1.5 nm of SiO₂ built from tridymite or a relaxed amorphous cell, about 150 to 250 atoms) with one Pb pointing into an oxide void, and final states at a back-bond centre, an oxide interstitial, and HPb + H → Pb + H₂ with a second H nearby. On one H100 an SCF of that size takes minutes; a 9-image NEB of 50 to 100 iterations is one to three days per path. (3) Prefactors: ph.x at the initial state and the saddle (harmonic transition state theory, Vineyard) gives kd0 to compare with 1.2×10¹² s⁻¹, and the same with deuterium masses gives the isotope effect Johnson 1981 needs. (4) The spread: the same path at several Pb sites of an amorphous interface gives a distribution of Ed to compare with σ = 0.105 eV. Each result enters the program as STATUS: computed and is scored against this page's 16 rows. ([Stesmans and Afanas'ev 1996](/writing/interface-study-stesmans-1996))
• Partial pressures above 1 atm. The paper oxidizes and passivates in 1.1 atm; the runtime refuses any partial pressure above 1 (mix:H2=1.1 is rejected). Allowing it would let the stated pressure through; in the equilibrium program it moves the uncapped share by a factor 1/√1.1. ([Stesmans and Afanas'ev 1996](/writing/interface-study-stesmans-1996))
• A vacuum with a pressure. “Diffusion-pumped vacuum (⩽4×10⁻⁷ Torr)” is written as vacuum. Johnson 1981's vacuum, which removes Pb where this one restores it, has no pressure at all. AMBIENT vacuum PRESSURE 4e-7 Torr and a residual-gas composition would let the two be told apart; no program reads either yet. ([Stesmans and Afanas'ev 1996](/writing/interface-study-stesmans-1996))
• One sample annealed in a series. The cycled sample is one slice taken through passivation and anneal at rising temperatures with ESR in between. Each history here is cut to its last cycle. A REPEAT … FOR T IN … with an observation after each pass would state the sequence and let a program with memory (created Pb) be tested on it. ([Stesmans and Afanas'ev 1996](/writing/interface-study-stesmans-1996))
• An observable named for what it is. The program's target is called the share left uncapped “at equilibrium”, and after a vacuum anneal it is not: it is the rate hypothesis. The predictor should report the equilibrium share (1 in vacuum, the bound in hydrogen) separately from the share at the end of the history, so a page cannot read one as the other. ([Stesmans and Afanas'ev 1996](/writing/interface-study-stesmans-1996))
• A SiC face vocabulary beyond Si, C and a: (03-38) is written here as 4H_SiC_03-38_face, a new word no program or ontology entry knows. Faces should be Miller indices (with polytype and off-cut), so a program can say which it has constants for and the runtime can refuse a misspelling. ([Yano et al. 2002](/writing/interface-study-yano-2002))
• The interface-state program's energies should be the sample's to choose (OBSERVE Dit AT Ec-0.15 eV), or the program should report the whole curve it can support; today only 0.3, 0.4 and 0.5 eV exist, and this paper's result lies at 0.1 to 0.2 eV. ([Yano et al. 2002](/writing/interface-study-yano-2002))
• The sic_interface_states program returns nothing (not even a refusal with a reason) when a history has no dry-oxygen step. A wet history should be refused with a stated reason, so the page can say why, rather than appear as “no program”. ([Yano et al. 2002](/writing/interface-study-yano-2002))
• In situ steps: a statement that two RUNs are in the same tube with no unload between them (IN_SITU), against Kobayashi's anneal in another furnace (a transfer through air). Here the argon anneal is in situ, and the gas switch in between is an unknown we had to write as an extra step. ([Yano et al. 2002](/writing/interface-study-yano-2002))
• The measurement temperature and method on an observable (MEASURE Dit METHOD conductance AT 300 K; C-V at 100 K): the same sample gives different counts of shallow states by method and temperature, which is how this paper argues. ([Yano et al. 2002](/writing/interface-study-yano-2002))
• OFFCUT read by the SiC growth program (constants per face and off-cut). Song's Si face is on axis, Goto and Hijikata's is 8° off, this chapter's is unrecorded, and the Si-face constants do not transfer between them (28 to 64% low here). The field exists on samples and no program reads it. ([Gupta and Akhtar 2011](/writing/interface-study-gupta-akhtar))
• A ramp rate as a statement (RAMP RATE 5 C/min FROM 800 TO 1150), so the chapter's stated rate is written as the paper gives it, not as a duration we derive. ([Gupta and Akhtar 2011](/writing/interface-study-gupta-akhtar))
• The sic_growth program's scope check labels a history outside its tested scope whenever any step is not dry oxygen, so a load and ramp in nitrogen, which grows nothing, marks every number here “outside the tested scope”. The check should look only at the steps that grow oxide. ([Gupta and Akhtar 2011](/writing/interface-study-gupta-akhtar))
• Bubbler temperature as an ambient statement (H2O FROM bubbler AT 85 C) that the runtime turns into a water partial pressure, so a wet recipe is written as the paper states it rather than as our computed mixture. ([Gupta and Akhtar 2011](/writing/interface-study-gupta-akhtar))
• A cumulative or replicate series: REPLICATES 4 on a measurement, so the four samples per point are part of the record (and their scatter, if a paper gives it, becomes the tolerance). ([Gupta and Akhtar 2011](/writing/interface-study-gupta-akhtar))
• A grid whose length is a state (GRID oxide { FROM: 0; TO: x_ox; CELLS: 50 }), or a stretched coordinate with position-dependent velocity, so interstitial and oxidant profiles can live in the growing oxide. This is the one change that would let the Si and C emission model be written as a candidate structure beside the linear-parabolic law. ([Goto and Hijikata 2016](/writing/interface-study-goto-2016))
• integral(expr) over a product of fields (the paper's ∫ κ C_Si C_O dx), not only integral(F) of one field. ([Goto and Hijikata 2016](/writing/interface-study-goto-2016))
• A growth rate against thickness as an observable (OBSERVE rate OF x_ox AT Thickness), so in-situ rate data can be scored directly instead of integrated into times that carry every reading error. ([Goto and Hijikata 2016](/writing/interface-study-goto-2016))
• A time origin at gas admission and an observation that starts later (MEASURE … FROM first_resolved_thickness): the paper's clock starts when argon is replaced by oxygen, its data at 3 to 13 nm; we had to anchor the clock with the program's own constants. ([Goto and Hijikata 2016](/writing/interface-study-goto-2016))
• OFFCUT read by the growth program: this Si face is 8° off axis, Song's on axis, and Song's Si-face constants are up to 36% slow here. The C face is 0.5° off. ([Goto and Hijikata 2016](/writing/interface-study-goto-2016))
• The sic_growth scope check marks a history outside its tested scope when any step is not dry oxygen, including an argon heat-up that grows nothing. It should check only the steps where oxide grows (and the temperatures of those). ([Goto and Hijikata 2016](/writing/interface-study-goto-2016))
• Nitric oxide (and N2O, NH3) as gases. The NO anneal at 1200 °C for 70 min is a stated step the language cannot write; it is an EVENT here. mix:NO=1.0 would let it be simulated by a program that reads p_NO (nitrogen at the interface, Figure 9; the interface-state reduction, Figure 6). ([Li, Luo and Ye 2023](/writing/interface-study-li-2023))
• An implant statement. IMPLANT P ENERGY 30 keV DOSE 1e13 [THROUGH FieldOxide] as a history step, carrying the species, the projected range and the damage a program can read. The paper's central variable is a sample field no program reads. ([Li, Luo and Ye 2023](/writing/interface-study-li-2023))
• A profile observable with the instrument in it. The composition target has no observation form: an OBSERVATION_MODEL returns one number. A profile result (value against position from the interface) and a reduction such as convolve(F, w) with a stated probe width, sampled at positions, would let Figure 8 be scored and would separate the instrument from the interface. ([Li, Luo and Ye 2023](/writing/interface-study-li-2023))
• A grid whose length follows a state (the oxide thickness), so a surface sink sits at the surface while the oxide grows; today the candidate program's sink is at a fixed 60 nm and its carbon is an upper bound. ([Li, Luo and Ye 2023](/writing/interface-study-li-2023))
• OFFCUT read by programs. The 4° off-cut is recorded and ignored; it sets the step density that roughens the interface the EDS line crosses, and Song's constants are on axis. ([Li, Luo and Ye 2023](/writing/interface-study-li-2023))
• A rate observable, measured at a thickness, not at a time. The chapter's data are dX/dt against X. The language observes states at the end of a history, so the rate had to be integrated to thickness against a reconstructed time with an unknown time zero. An observation of a state's time derivative (EQUATION: Growth_rate = d(x_ox)/dt) and a result indexed by a state (MEASURE Growth_rate AT Thickness = 1.0e-6) would compare the rates directly and need no time zero; it would also have answered “is the initial rate enhanced?” in one row. ([Hijikata, Yagi, Yaguchi and Yoshida 2013](/writing/interface-study-hijikata-2013))
• A measurement during a RUN (MEASURE <observable> DURING <run> EVERY <s> or in situ), so an in-situ ellipsometry run is one history with many measured points instead of a curve sampled by re-running the history to each time. ([Hijikata, Yagi, Yaguchi and Yoshida 2013](/writing/interface-study-hijikata-2013))
• Clock reference (as for Song 2004): TIME_ZERO at_gas_on | at_first_point, so an in-situ run's time axis can be stated rather than searched. ([Hijikata, Yagi, Yaguchi and Yoshida 2013](/writing/interface-study-hijikata-2013))
• Predictor: a thin-oxide channel in the SiC growth program, with its own pressure law. The Si and C emission model is writable today: FIELDs for Si and C interstitials and oxygen in the oxide on a grid travelling with the interface, FLUX with D_Si, D_C, D_O, BOUNDARY fluxes ν_Si k C_O, ν_C k C_O and (2 − ν_Si − ν_C − α/2) k C_O at the interface (Eq. 10), an OxideGrowth rate k0 (1 − left(C_Si)/C_Si⁰)(1 − left(C_C)/C_C⁰) C_O / N0 (Eq. 8). Table 1 gives the oxide-side constants; ν_Si, ν_C and k0 per face and temperature exist only as points on Figure 6, so they would be graph reads. Its grid would need to follow the growing oxide (no grid of varying length today). ([Hijikata, Yagi, Yaguchi and Yoshida 2013](/writing/interface-study-hijikata-2013))
• Predictor: the program scales its whole rate by p_O2. Hijikata's initial rate R_r is nearly independent of pressure while R_g is proportional (Figure 8). Any SiC program used for low-pressure or ramp-in-diluted-oxygen histories needs the two separated. ([Hijikata, Yagi, Yaguchi and Yoshida 2013](/writing/interface-study-hijikata-2013))
• One oxide per history, from a growth law for the gas. Two programs grow 134 and 37 nm on the same 150 min history (59 nm measured); the inventory's germanium dose follows its own oxide. The predictor should take one oxide thickness (or a stated one: MEASURED_OXIDE_NM as an input, as done for the nanowire program's RUN_FLAT_OXIDE_NM) and feed it to every composition program. With the stated oxide the inventory program's width is a prediction of Long's profiles with nothing grown. ([Long et al. 2012](/writing/interface-study-long-2012))
• A dry-oxygen SiGe growth program, or a refusal. The Rabie program runs dry histories at its wet 800 °C constants and labels them “direction only”. Long gives eleven dry oxide thicknesses at 800 to 1000 °C on two compositions; they would fit or test one. ([Long et al. 2012](/writing/interface-study-long-2012))
• STARTING_OXIDE_NM read by the SiGe programs. The native oxide is 5 nm here (a quarter of the 25 min oxide), and it carries GeO2 (XPS). Neither SiGe program reads it. ([Long et al. 2012](/writing/interface-study-long-2012))
• Layered sample with a finite alloy. LAYER SiGe { COMPOSITION: Si0.85Ge0.15; THICKNESS: 80 nm } on LAYER Si { THICKNESS: 100 nm }: needed to say when the pile-up consumes the layer (Long's Figure 6 at 150 min, and his Eq. 8). ([Long et al. 2012](/writing/interface-study-long-2012))
• An observation model for the instrument's depth resolution. RBS and XPS report a profile smoothed by about 5 nm (estimated here from the un-oxidized edge) and 3.5 or 11.7 nm (Ge2p, Ge3d). A statement such as OBSERVATION_MODEL ... RESOLUTION gaussian 5 nm would let a sharp predicted profile be compared with a measured one without the hand-written smoothing in this study file. ([Long et al. 2012](/writing/interface-study-long-2012))
• A depth profile as a target. Results are scalars at chosen depths (10 nm, 30 nm). Long's data are profiles with the front aligned at zero by the authors; a profile target (depth series, with the front position as an unknown) would score the whole shape. ([Long et al. 2012](/writing/interface-study-long-2012))
• The conserved dose in the inventory program. Its comment follows Long's Eq. 7 (the germanium of the consumed alloy, x N_ox X/(1−X)) but places that dose as the excess above X in an exponential. Counting atoms with the alloy shrinking by the silicon removed gives an excess of x N_ox X, as scripts/analyse_germanium_inventory.py already uses. The program's width is 1/(1−X) too large: 18% at X = 0.15, 100% at X = 0.5. A predictor change, not a language one; it is listed here because this paper is its source. ([Long et al. 2012](/writing/interface-study-long-2012))
• A clean word for “native oxide not removed”. The CLEAN vocabulary has no entry for an uncleaned surface; not_stated is wrong here, because the paper does state it. ([Long et al. 2012](/writing/interface-study-long-2012))
• Interface-resident species and operators that act along an interface. A FIELD on an interface (surface concentration of mobile germanium) with Transport along it, a source tied to the oxidation flux, and an exchange into the crystal below. Without it, the paper's mechanism and every one of its 14 measured lengths and widths cannot be written. ([Thornton et al. 2022](/writing/interface-study-thornton-2022))
• A second spatial coordinate, or a sidewall coordinate. GRID is one-dimensional and travels with a planar front; a fin needs height along the sidewall and lateral depth. ([Thornton et al. 2022](/writing/interface-study-thornton-2022))
• SHAPE read by every program, with a refusal. The SiGe programs check only nanosheet and nanowire; a SHAPE: fin sample is run as a planar alloy without a label. Every program should state which shapes it accepts and label the rest outside its scope. ([Thornton et al. 2022](/writing/interface-study-thornton-2022))
• Layered samples. LAYER blocks (composition, thickness, strain) for a superlattice on a silicon base, so that a program knows each SiGe layer holds 15 nm of alloy. ([Thornton et al. 2022](/writing/interface-study-thornton-2022))
• A contour observable. OBSERVATION_MODEL ... EQUATION: L = extent(x_Ge > 0.03 along sidewall): an observable defined by a level of a field. ([Thornton et al. 2022](/writing/interface-study-thornton-2022))
• Residual gas of a sealed ampule. The ambient was written as air at 35 mtorr; a statement for “sealed, pumped to P, gas unknown” with the oxygen partial pressure as an unknown (VARY p_O2) would let alignment search it. ([Thornton et al. 2022](/writing/interface-study-thornton-2022))
• Sidewall orientation. ORIENTATION is a wafer property; a fin's sidewall plane is a separate property that the growth programs should read. ([Thornton et al. 2022](/writing/interface-study-thornton-2022))
• Which step is the oxidation. A lamp heat-up written in oxygen, as an RTP usually runs, makes the trap-density program read the oxidation temperature as 25 °C and refuse every history here; the heat-up had to be written in vacuum, which removes 0.2 to 1.6 nm of oxide the programs would otherwise grow. A ROLE oxidation|ramp|anneal|pull on each RUN would let programs find the step by what it is. ([Fukuda et al. 1992](/writing/interface-study-fukuda-1992))
• Isotopes and process gases. Deuterium had to be written as hydrogen (pure_H2), silane as nitrogen, and the phosphorus source not at all: the language's gases are Ar, H2, H2O, N2 and O2. mix:D2=1 and mix:SiH4=…, POCl3 would record what was done; a program could then ignore them, or carry an isotope effect (the paper's SIMS counts deuterium). ([Fukuda et al. 1992](/writing/interface-study-fukuda-1992))
• A deposition and a doping as steps. The 350 nm polysilicon at 620 °C and its 900 °C doping are written as nitrogen holds. A DEPOSIT polysilicon 350 nm statement that caps the oxide (no gas reaches it afterwards) would say why the gas does not matter and let a program ask whether hydrogen can leave through the cap, which is what the computed bound below depends on. ([Fukuda et al. 1992](/writing/interface-study-fukuda-1992))
• An observation's timing. The trap density and charge were measured after the gate steps, not after the anneal the figure is about. The language ends the history at the measurement implicitly. An explicit OBSERVE Dit, Nf AFTER <step> would make the gap between the anneal and the measurement visible on the page and let a program answer at the anneal as well. ([Fukuda et al. 1992](/writing/interface-study-fukuda-1992))
• Thickness and time on separate figures. Figures 2 and 3 are the same wafers against thickness and against time, never paired. A sample field OXIDE_NM read as a measured property beside a stated time would let each wafer carry both, and let a growth program be scored on it without a pairing of ours. ([Fukuda et al. 1992](/writing/interface-study-fukuda-1992))
• Where the hydrogen goes. Not a language gap but a program one the language exposes. Since commit 7f1ec87 the trap-density program's capping has Brower's activation energy and capped sites release hydrogen at 900 °C, but released hydrogen leaves the model: nothing holds the deuterium SIMS finds in the oxide (1.7×10^12 to 2.8×10^13 cm⁻²), so nothing can recap on cooling under the polysilicon, and the 900 °C gate doping erases every anneal. A hydrogen inventory per layer, and a DEPOSIT that caps it, would let a program ask what the paper's wafers kept. ([Fukuda et al. 1992](/writing/interface-study-fukuda-1992))
• Which step is the oxidation. Writing the push in oxygen, which may be what was done, makes the trap-density program read the oxidation temperature as 25 °C and refuse. A ROLE oxidation|load|pull|anneal on each RUN would let a push be written as the heating it is. ([Kar and Narasimhan 1987](/writing/interface-study-kar-1987))
• An etch that removes oxide. The 1100 °C sacrificial oxide is grown and stripped in HF before the gate oxide; it had to be an event. ETCH REMOVE ALL (or TO <nm>) after a RUN would let the preoxidation be simulated, so that its heat reaches the programs and its oxide does not. ([Kar and Narasimhan 1987](/writing/interface-study-kar-1987))
• Dit(E) as an observable. The predictor's energy rows are fixed target names written only after hydrogen. An observation Dit AT_ENERGY <eV> ABOVE Ev BY admittance|quasistatic with the substrate type would let a paper's distribution be stated as what was measured, and let a program say it cannot answer at that energy rather than be silent. ([Kar and Narasimhan 1987](/writing/interface-study-kar-1987))
• Peaks and cross sections. The paper's central results are a peak's height and energy and a capture cross section against thickness. The ontology has no observable for any of them. ([Kar and Narasimhan 1987](/writing/interface-study-kar-1987))
• A device tied to a calibration series. Each measured device is known by its thickness; its oxidation time comes only from Figure 1. A sample field OXIDE_NM read as a measured property, and a way to say a RUN's duration is whatever makes that thickness (an inverse VARY), would replace the times inferred by hand here. ([Kar and Narasimhan 1987](/writing/interface-study-kar-1987))
• Doping and gate metal as inputs. DOPING is recorded and read by no program; here it decides which half of the gap was measured. The gate metal (Au on n-type, Al on p-type) is not recordable; an aluminium gate can release hydrogen at the interface even without an anneal. ([Kar and Narasimhan 1987](/writing/interface-study-kar-1987))
• A clean word. The vocabulary has no sacrificial_oxide_HF for “preoxidation was part of the surface cleaning”; CLEAN had to be not_stated although the paper states it. ([Kar and Narasimhan 1987](/writing/interface-study-kar-1987))
• Ammonia and nitrous oxide as gases. AMBIENT mix:NH3=1.0 and mix:N2O=1.0 are refused (the gases are Ar, H2, H2O, N2, O2). Adding p_NH3 and p_N2O as controls would let a nitridation be written as done; today it can only be written as nitrogen, which carries its heat and nothing else. Every row here would then at least be run on the paper's history. ([Tsai, Wu, Wu and Hu 1987](/writing/interface-study-tsai-1987))
• A program for nitrogen at the interface. With the gas written, a program has to say what it does: incorporate nitrogen at the Si/SiO2 interface (rate rising with temperature, saturating with time), release hydrogen into the film, and shift fixed charge and trap generation. The paper's Table II (more negative flat-band voltage at lower nitridation temperature, positive after 3 h at 1100 and 1150 °C) is the first check such a program would have. ([Tsai, Wu, Wu and Hu 1987](/writing/interface-study-tsai-1987))
• An electrical stress step. Figure 1 and 2 are measured during constant-current injection; the language has no statement for it, so the paper's main result is outside any history. A STRESS statement with current density, polarity and fluence, and observables for generated trap density and flat-band shift, would make it a target. ([Tsai, Wu, Wu and Hu 1987](/writing/interface-study-tsai-1987))
• The gate as a sample property a program reads. A flat-band voltage needs the gate work function (As-doped polysilicon here). GATE is recorded on the sample and read by nothing; a program computing Vfb from fixed charge, trapped charge and the gate would let Table II be scored. ([Tsai, Wu, Wu and Hu 1987](/writing/interface-study-tsai-1987))
• The forming-gas time and H2 fraction (history). Since 11 November the capping rate has Brower's 1.66 eV (see the Reed 1989 page), so a forming-gas ramp from room temperature counts for little (the cold-furnace ramps lower the density 37%, where they gave ten times before). What still sets the predicted density is what the paper does not state: the forming-gas time (10 to 60 min) and the H2 fraction (5% gives 3.7 times the density of 10%). ([Tsai, Wu, Wu and Hu 1987](/writing/interface-study-tsai-1987))
• Process steps that bring hydrogen. LPCVD polysilicon (silane), CVD oxide and the ammonia itself are hydrogen sources the history records only as events. A hydrogen source on an EVENT or a deposition RUN with its gas would let them reach the trap-density program. ([Tsai, Wu, Wu and Hu 1987](/writing/interface-study-tsai-1987))
• Hydrogen release tied to what the history put there (predictor). The program now releases hydrogen (Si–H → Si• + H at Brower's rate), but a release can only raise the density where an earlier step capped sites, and on this page none can. It also has no release for the second population and does not use the computed Si–H energy against the gas's p_H2. Until the hydrogen sources below can be written, an RTA, a vacuum anneal and a rapid pull still look the same as no step. ([O'Sullivan et al. 2001](/writing/interface-study-osullivan-2001))
• Hydrogen brought by a step that is not a gas anneal. Polysilicon from silane at 620 °C, POCl3 doping, plasma steps and moisture during a slow pull passivate the interface (the paper, PDF p. 9-10). The language records them as EVENTs that do nothing. A SOURCE hydrogen clause on a deposition EVENT, or an AMBIENT for a deposition (mix:SiH4=…), and a stated moisture on pulls, would let a program see them. The vocabulary of gases is Ar, H2, H2O, N2, O2; silane, POCl3, NH3 and N2O are refused. ([O'Sullivan et al. 2001](/writing/interface-study-osullivan-2001))
• Which step is the oxidation. The RTO heat-up had to be written in nitrogen: the trap-density program takes the oxidation's temperature and time from the first dry-oxygen RUN, so a heat-up in oxygen from 25 °C would be read as an oxidation at 25 °C. A ROLE oxidation|anneal|ramp|pull on each RUN (as Hung's page asked) would remove the trap. ([O'Sullivan et al. 2001](/writing/interface-study-osullivan-2001))
• A gate on the sample. Polysilicon, aluminium and a mercury probe behave differently under hydrogen (the paper: lightly doped polysilicon needs a longer alloy). GATE is recorded here as a sample field no program reads. A gate field in the ontology, read by the hydrogen operator as a barrier or a source, would let processes 1-2 and 3-5 differ. ([O'Sullivan et al. 2001](/writing/interface-study-osullivan-2001))
• A pull as a speed. “a pull from the furnace at 600 °C at 8 cm/min” is a speed, not a duration or a cooling rate; the boat travel is unknown. A PULL SPEED 8 cm/min FROM 600 statement with a furnace profile would turn it into a temperature against time. ([O'Sullivan et al. 2001](/writing/interface-study-osullivan-2001))
• A measurement that is a difference. Table I's peaks are densities after subtracting a background measured on another (hydrogen-passivated) sample. An observation statement naming the reference sample would let such numbers be compared with a difference of two simulations. ([O'Sullivan et al. 2001](/writing/interface-study-osullivan-2001))
• An HF dip as a clean. CLEAN accepts named cleans only; a bare pre-oxidation HF dip (process 3) had to be written not_stated. ([O'Sullivan et al. 2001](/writing/interface-study-osullivan-2001))
• A gate that supplies hydrogen. Reed's Al-gate capacitors anneal at the same rate in any ambient because aluminium reacting with hydroxyl at the Al–SiO2 interface releases atomic hydrogen. The language can record GATE: Al on a sample, but no statement says the gate is a hydrogen source during every later heat; an RTA in nitrogen therefore anneals nothing. A sample layer with a declared reservoir (LAYER Al { SOURCE: H }) or a SOURCE H clause on a RUN would let the Al-gate curves be simulated at all. ([Reed 1989](/writing/interface-study-reed-1989))
• Atomic against molecular hydrogen. Bare oxides anneal much faster in atomic hydrogen than in H2 (PDF p. 3); the language's gases are Ar, H2, H2O, N2 and O2. An H species in mix: (and a program that reads it) is needed before corona, plasma or post-metal anneals can be written. ([Reed 1989](/writing/interface-study-reed-1989))
• A measured starting state. Every case here starts from an oxidation we invented because the program cannot be told “the sample has 8.4×10^10 cm⁻² eV⁻¹ of traps at t = 0”. A sample field such as STARTING_DIT: 8.4e10 (as STARTING_OXIDE_NM does for thickness) would let an anneal series be tested on its kinetics alone, without an unknown oxidation in front of it. ([Reed 1989](/writing/interface-study-reed-1989))
• One sample measured along an anneal. Each point of a decay curve is written as a separate history ending at that time. Whether the paper annealed one capacitor repeatedly (with cooling and a measurement between anneals) or a fresh one each time is not stated, and the language cannot say which. A REPEAT of anneal-and-measure on one sample, with the cooling between, would make the difference explicit. ([Reed 1989](/writing/interface-study-reed-1989))
• A plateau in the trap-density program (a predictor change). Commit 7f1ec87 added Brower's Arrhenius capping (1.66 eV) and dissociation of Si–H (2.56 eV); the steady state they give is 18 to 110 times below Burte's plateau. Still missing: a floor of sites hydrogen does not cap, or a spread of capping barriers, and a rate level set by data at more than one temperature. ([Reed 1989](/writing/interface-study-reed-1989))
• An etch that removes oxide. The 1100 °C preoxide is grown and stripped before the oxidation that matters. The language has no statement that removes oxide, so the step had to become an EVENT and its 30 min at 1100 °C left out of the simulated heat; written as a RUN it stays as about 70 nm of oxide under the real one. ETCH REMOVE ALL (or ETCH TO 0 nm) would let the strip be simulated and the heat kept. ([Kar, Shanker and Chari 1985](/writing/interface-study-kar-1985))
• Which RUN is the oxidation. The trap-density program takes the oxidation temperature from the first dry-oxygen RUN. A sacrificial oxidation in front of the real one changes the predicted density by a factor of 2.7 (4.3 to 1.6 × 10¹¹) through that choice alone. A ROLE sacrificial|oxidation|anneal|pull on each RUN would let programs find the oxidation by what it is (the same gap Hung 1987 hit with a boat-in). ([Kar, Shanker and Chari 1985](/writing/interface-study-kar-1985))
• A measured thickness as a constraint on an unstated time. The paper states thicknesses, not times. Here the time was set by hand from the thickness and thickness was then dropped from the score, so the search could double the time unchecked. A sample field the alignment reads as a constraint (OXIDE_NM_MEASURED checked against the growth program while other quantities are searched) would pin the time and say so. ([Kar, Shanker and Chari 1985](/writing/interface-study-kar-1985))
• A band-edge peak as an observable. Every number in Figure 2 is the height of a peak near Ev or Ec of an unannealed distribution. The observable vocabulary has midgap density and, after hydrogen only, a fixed set of energies. An observable Dit_peak_near_Ev / Dit_peak_near_Ec with its energy, and a program that predicts a distribution without hydrogen, are needed before any of the 18 peak values can be compared. ([Kar, Shanker and Chari 1985](/writing/interface-study-kar-1985))
• The clean. “Degreasing in hot solvents, ultrasonic cleaning, and etching in acids” followed by a sacrificial oxide has no CLEAN word in the vocabulary; it is written not_stated. A CLEAN: sacrificial_oxide_HF would record the most consequential part of it. ([Kar, Shanker and Chari 1985](/writing/interface-study-kar-1985))
• How the measurement was made. The densities are from illuminated admittance (capacitance and conductance, 30 Hz to 100 kHz) on 100 Å semitransparent Au or Al dots. Nothing in a history or result says which technique gave a density, and techniques disagree near the band edges; an observation statement with the method would let results from different methods be kept apart. ([Kar, Shanker and Chari 1985](/writing/interface-study-kar-1985))
• A freezing point that is stated and testable. The as-grown series is the cleanest test of where the oxygen exchange stops: ten growth temperatures, one pull. With the stated hypothesis at 350 °C a realistic cool erases the growth temperature; at 450 °C it does not. The ontology could carry the freeze-out temperature of each operator as a declared hypothesis with its own test set (this series and Deal and Razouk's pulls), so that raising it is a recorded decision, not a template edit. ([Akinwande, Ho and Plummer 1981](/writing/interface-study-stanford-1981))
• The gas a wafer cools in. A fast pull leaves the oxidation gas for a flushed mouth and then room air. The best variant here (cooling in nitrogen) is a Python edit; a PROFILE or WITHDRAW statement that names the gas after withdrawal would make it a declared unknown (VARY ambient). ([Akinwande, Ho and Plummer 1981](/writing/interface-study-stanford-1981))
• A starting charge from the laboratory. The program's as-grown value is Deal's Fairchild line. INITIAL Q_f FROM <laboratory line> would let Stanford's own as-grown line (this report's Figure 7.7) be the start for Stanford's wafers, and make the Fairchild line one stated choice. ([Akinwande, Ho and Plummer 1981](/writing/interface-study-stanford-1981))
• A target observed on a fixed time grid, not a history. The argon anneal times were “adjusted at each temperature to reach an equilibrium value” (1983 report). The language can only write a guessed duration. RUN … UNTIL steady (an anneal that lasts until the state stops moving) would state what the authors did and make the anneal time a program output instead of an assumption. ([Akinwande, Ho and Plummer 1981](/writing/interface-study-stanford-1981))
• One wafer measured between anneals. The cycling is one sequence with a C-V after each step; it is written as five histories. A MEASURE event inside one history whose readings are the scored values would say it once, and would show that metal must come off between steps. ([Akinwande, Ho and Plummer 1981](/writing/interface-study-stanford-1981))
• Oxidation rate as a sample-history quantity. Eq. 7.1 makes the as-grown charge a function of dx/dt at the end of growth. The growth programs compute it; the charge program cannot read another program's output. A cross-program QUANTITY (the oxidation rate at the pull) would let a rate-dependent charge law be written and tested on Figure 7.7. ([Akinwande, Ho and Plummer 1981](/writing/interface-study-stanford-1981))