{"id":"dc620b65-0963-4796-a665-7bd8ccbbcc6b","arxiv_id":"2502.02714","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A 191 nm gate beta-Ga2O3 HFET reaches fT=32 GHz, fMAX=65 GHz, gm=110 mS/mm, and shows no current collapse after Al2O3 passivation.","lead":"This paper reports a gallium oxide transistor that reaches a 32 GHz current gain cutoff and a 65 GHz power gain cutoff, among the best reported for this material. The result matters because gallium oxide is a candidate for next-generation high-power RF amplifiers and switches, and the paper also shows a passivation layer that removes current collapse.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"RF figures rely on open-only de-embedding and unshown extrapolation; raw data are needed to verify fT=32 GHz and fMAX=65 GHz.","rationale":"The paper's title and abstract foreground fT=32 GHz and fMAX=65 GHz, making the RF extraction the pivotal quantitative result. The method description is sparse: it mentions an open-pad de-embedding and extrapolation to 0 dB, but provides no raw S-parameter figures, no extrapolation slope, and no comparison of measured versus de-embedded gain. The reader's weakest assumption correctly targets this gap. I considered whether the passivation comparison is a more load-bearing issue, since the 'no current collapse' claim uses different quiescent biases before and after deposition, but that claim is secondary to the headline RF numbers and is also undermined by the lack of raw data. The proposed concrete test—obtaining and independently re-analyzing the S-parameters—would directly settle whether the reported fT/fMAX are reliable. The paper is otherwise a competent process demonstration, and the DC results support the low-interface-resistance improvement, so the conditional verdict is appropriate.","tokens_in":9291,"tokens_out":10476,"duration_ms":91044,"concrete_test":"Request the raw S-parameters or the post-de-embedding h21 and U traces from the authors. Re-extract fT and fMAX by fitting a -20 dB/dec slope over at least one decade of data (e.g., 2-20 GHz) and compare with the reported values. If the fitted roll-off deviates from -20 dB/dec by more than 2 dB/dec, or if re-analysis with open-short de-embedding changes fT/fMAX by more than 10%, the headline numbers are not robust. Additionally, request pulsed IV data on the same device before and after Al2O3 passivation with VGS,q = VTH - 2 V and VDS,q = 0 and 5 V to verify the no-collapse claim under controlled bias conditions.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claims (fT=32 GHz, fMAX=65 GHz, fT·LG=6.1 GHz·µm) rest on S-parameter measurements from 100 MHz to 20 GHz, de-embedded with an isolated open pad only, and extrapolated to unity gain without showing raw gain traces, extrapolation lines, or measured versus de-embedded gain. If the gain roll-off deviates from -20 dB/dec, or if open-only de-embedding leaves series pad resistance and inductance in the data, the extracted fMAX (extrapolated about 10 dB above 20 GHz) and fT could shift substantially. This is the most load-bearing concern because the headline advance is the RF performance; the passivation claim also depends on RF gain that appears only after Al2O3 deposition, but the before/after passivation comparison uses different quiescent gate biases relative to threshold (VGS,q=-4 V before, VGS,q=-8 V with VTH=-6 V after), so the attribution is not fully controlled without raw pulsed waveforms.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This letter reports a β-(AlxGa1-x)2O3/Ga2O3 heterostructure FET fabricated with low-power BCl3/Ar and SF6/Ar plasma etching, an HCl surface clean, and ultra-high-vacuum annealing before MBE regrowth of degenerately doped source/drain contacts. The device is reported to achieve DC ID = 0.5 A/mm, RON = 6.1 Ω·mm at VGS = 3 V, peak gm = 110 mS/mm, fT = 32 GHz, fMAX = 65 GHz, and an fT·LG product of 6.1 GHz·µm for LG = 191 nm. The authors further claim that 100 nm Al2O3 passivation eliminates current collapse, based on pulsed I-V measurements before and after passivation and on the appearance of RF gain only after passivation.","tokens_in":9485,"tokens_out":6383,"duration_ms":58596,"significance":"If the reported RF numbers are confirmed, this is a meaningful advance for β-Ga2O3 RF transistors: the fT·LG product and fMAX are among the highest reported for the material system, and the process improvements (low-power etch, HCl clean, vacuum annealing before regrowth) appear to reduce the regrowth interface resistance that limited earlier devices. The central DC and RF figures are direct measurements rather than outputs of a fitted model, and the fT·LG product follows from measured fT and LG without free parameters; the cryogenic characterization and TLM analysis are additional strengths. However, the verification of the headline RF claims is currently incomplete: no raw S-parameter traces or extrapolation lines are shown, the de-embedding is open-pad-only, and the before/after passivation comparison uses different quiescent bias conditions. These gaps are load-bearing because the novelty of the letter rests on the RF performance and on the passivation claim.","major_comments":[{"comment":"The headline fT = 32 GHz and fMAX = 65 GHz are obtained by extrapolating |h21| and U to 0 dB from measurements taken only up to 20 GHz, but the manuscript does not show the measured gain traces, the extrapolation lines, or a comparison of raw and de-embedded data. Since fMAX = 65 GHz lies well above the 20 GHz measurement ceiling, this result depends entirely on the assumed -20 dB/dec roll-off and on the adequacy of the open-pad-only de-embedding described in the RF paragraph and Ref. 28. Open-only de-embedding leaves series pad resistance and inductance in the data, which can shift both fT and fMAX. Please provide the raw S-parameters, the de-embedded |h21|, U, and MAG/MSG curves with extrapolation lines for the bias points of Fig. 4(a)-(c), and state the resulting uncertainty in the extrapolated fT and fMAX.","section":"RF measurements, Fig. 4"},{"comment":"The before/after passivation comparison is not fully controlled. Before passivation, the gate-lag and double-pulse quiescent biases are VGS,q = -4 V, while after passivation the same measurements use VGS,q = -8 V, justified by VTH = -6 V for that device; the pre-passivation threshold voltage is not reported. If VTH differed before passivation, the two measurements do not sample the same trap-filling condition, so the attribution of the disappearance of current collapse to Al2O3 is not conclusive. In addition, the statement that the device showed no RF gain before passivation (RF paragraph) is made without showing the corresponding RF traces, even though DC-RF dispersion is invoked as the cause.","section":"Pulsed IV and passivation, Fig. 3"},{"comment":"The record claims are based on different best devices for different metrics: ID = 0.5 A/mm and RON = 6.1 Ω·mm are from a 173 nm gate device, peak gm = 110 mS/mm is from a 100 nm gate device at VDS = 15 V, fT = 32 GHz is from a 191 nm gate device, and fMAX = 65 GHz is obtained at VDS = 21 V. No error bars, device-to-device spread, or wafer statistics are reported, and the fT·LG benchmark in Fig. 4(d) uses the single highest product 'among other values'. The state-of-the-art conclusions would be substantially strengthened by reporting the range and median of these key figures across the measured devices and by using a consistent device-selection rule for all benchmark plots.","section":"Device benchmarking, Figs. 2(e), 4(d), 5"}],"minor_comments":[{"comment":"There are several typos and notation issues: 'theoritical' in the Supplementary Material, 'schotty' in the fabrication text, 'I shaped gate' should be 'I-shaped gate', and 'AlGaO/GaO HFET' should be 'AlGaO/Ga2O3 HFET' in several places. Also, '2500C' should be '250 °C', and the gate-length range '100-200 nm' is inconsistent with the specific 'DUT of 173 nm gate length (LG)' used for the main DC figures.","section":"Fabrication and notation"},{"comment":"The device names A and B are used without being formally defined; the reader cannot tell which device produced each of the reported ID, gm, fT, and fMAX values. Please label the devices consistently and specify which gate length and bias point correspond to each reported record.","section":"Device labels and measurement conditions"},{"comment":"The text states that fT·LG = 6.1 GHz·µm corresponds to Vsat ≈ 3 x 10^6 cm/s. Using the standard relation v_eff = 2π fT LG gives approximately 3.8 x 10^6 cm/s for 32 GHz and 191 nm; please state the formula and any correction factor used, or adjust the stated value.","section":"Saturation velocity estimate"},{"comment":"The claim that this is the first demonstration of successful Al2O3 passivation for β-Ga2O3 should be qualified relative to Ref. 9, which reports a similar Al2O3 passivation technique with moderate dispersion; a quantitative criterion for 'successful' (e.g., percentage of current collapse before and after) would make the novelty claim precise.","section":"Passivation novelty claim"}],"recommendation":"major_revision","confidential_remarks":"The manuscript reports promising device results, but the RF extraction transparency is below the standard I would expect for a headline frequency claim. I would make acceptance conditional on the authors providing raw S-parameter plots with de-embedded gain curves and extrapolation lines, and on addressing the passivation comparison bias-point mismatch. The stress-test concern about open-pad-only de-embedding is real and should be explicitly answered. I did not find circularity: the fT·LG product is a direct calculation from measured fT and LG. If the raw RF data confirm the extrapolations, this would be a solid Applied Physics Letters contribution; the current version is not yet verifiable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a competent device paper that likely does set a new mark for beta-Ga2O3 RF HFETs, but the reporting is a bit too tidy. I'd want the raw S-parameter traces and closer before/after passivation data before putting real weight on fT=32 GHz and fMAX=65 GHz.\n\nWhat's new: a combined regrowth process fix — low-power BCl3/Ar then SF6/Ar etching, HCl surface clean, and UHV anneal before MBE regrowth — that gives lower interface resistance and kills the low-VDS nonlinearity seen in their earlier HFETs. The DC numbers (0.5 A/mm, RON 6.1 Ω·mm, gm 110 mS/mm) and RF numbers (fT 32 GHz, fMAX 65 GHz, fT·LG 6.1 GHz·µm) are among the best reported for beta-Ga2O3. The Al2O3 passivation with no current collapse under 200-ns pulses is a first for this material system as far as I know; they cite Dryden et al. seeing only moderate dispersion, so this is a genuine data point. The claim that Al2O3 improves RF gain (they got no gain before passivation due to dispersion) is consistent.\n\nSoft spots, in order of bother:\n\n1. RF extraction. They use an open-pad de-embedding (Koolen method cited) and extrapolate h21 and U to 0 dB from a 100 MHz–20 GHz measurement without showing raw gain traces, extrapolation lines, or de-embedded versus measured response. fMAX=65 GHz means extrapolating about 10 dB past the last measurement point, so the -20 dB/dec assumption and residual pad inductance matter a lot. This is the load-bearing number, and it's under-reported.\n\n2. Passivation comparison. The pulsed-IV before/after uses VGS,q=-4 V on a device with VTH≈-4 V, and after passivation VGS,q=-8 V on a device with VTH=-6 V. That's a different overdrive below threshold, so it's not a clean apples-to-apples comparison. The after condition is actually more negative, which is a harsher test, so the direction favors their claim, but without raw waveforms or a same-device measurement the attribution is weaker than stated.\n\n3. Statistical reporting. We get best-device values (e.g., gm=110 mS/mm from a 100-nm device, fMAX at VDS=21 V) with no error bars or device-to-device spread. A letter format explains some of this, but it should at least show the RF plots.\n\nNo fitted parameters, no circular modeling, citation pattern is fine. This is a straightforward experimental advance.\n\nI'd send this to review. It's exactly the kind of incremental but concrete progress a field like Ga2O3 RF needs. Ask the referee to require the raw S-parameters and the before/after pulsed IV under matched quiescent conditions.","headline":"Solid beta-Ga2O3 HFET process-and-passivation demonstration with plausibly record RF numbers, but unshown raw RF extraction and a not-fully-controlled passivation comparison keep me from endorsing the exact figures.","tokens_in":10046,"tokens_out":3216,"would_cite":true,"duration_ms":29378,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A beta-(AlGa)2O3/Ga2O3 HFET with low-damage regrowth and Al2O3 passivation reaches 32 GHz current-gain and 65 GHz power-gain cutoff frequencies with no current collapse.","keywords":["beta-Ga2O3","HFET","RF cutoff frequency","Al2O3 passivation","MBE regrowth","current collapse","plasma etching","wide bandgap semiconductor"],"falsifier":"A direct test is to measure S-parameters up to at least 65 GHz on the passivated device and locate the unity-gain crossings of |h21| and unilateral gain without relying on extrapolation; if the measured rolloff deviates from 20 dB/dec or the crossings are below 32 and 65 GHz, the headline numbers would need revision.","tokens_in":9108,"feed_emoji":"⚡","tokens_out":5920,"duration_ms":47871,"temperature":0.7,"pith_summary":"This paper reports a beta-(Al,Ga)2O3/Ga2O3 heterostructure field-effect transistor whose fabrication was changed at two critical steps: the AlGaO barrier and Ga2O3 cap are removed with low-power plasma etches, and the wafer is cleaned and vacuum-annealed immediately before MBE regrowth of the degenerate N++ source/drain contacts. The authors argue that these changes lower the interface resistance between the regrown contacts and the two-dimensional electron gas, eliminating the low-voltage nonlinearity seen in their earlier devices. The resulting device delivers 0.5 A/mm drain current, 110 mS/mm peak transconductance, 6.1 Ω·mm on-resistance, and RF cutoff frequencies f_T = 32 GHz and f_MAX = 65 GHz, the latter among the highest reported for Ga2O3. After a 100 nm Al2O3 passivation layer is deposited, pulsed measurements show no current collapse, which the paper presents as the first successful Al2O3 trap passivation for beta-Ga2O3 devices. If these results hold, they strengthen the case for Ga2O3 as a candidate material for high-power RF amplifiers.","feed_headline":"Vacuum-annealed Ga2O3 HFET reaches 32/65 GHz","feed_subtitle":"A new regrowth process and Al2O3 passivation push beta-gallium-oxide RF transistors to record speeds with no current collapse.","key_machinery":"The load-bearing mechanism is the regrown N++ contact/2DEG interface. The paper reduces interface resistance by using low-power plasma etches (ICP 300/RIE 50 for BCl3/Ar; ICP 150/RIE 50 for SF6/Ar, etching at 1–2 nm/min), removing surface contaminants with a 1:3 HCl:DI water dip, and annealing at 600 °C in ultra-high vacuum ($10^{-10}$ torr) immediately before ozone-MBE regrowth of the N++ layer. This sequence is credited with eliminating the low-bias output nonlinearity of the earlier generation device and enabling the high gm and current. The other central element is the 100 nm Al2O3 passivation deposited by ALD at 300 °C, which the paper reports as passivating traps and eliminating DC-RF dispersion as seen in 200 ns pulsed I-V.","core_discovery":"The central claim is that a specific process sequence—low-power BCl3/Ar etching of the AlGaO barrier, very low-power SF6/Ar etching of the Ga2O3 cap, a 1:3 HCl dip, and a 600 °C anneal at $10^{-10}$ torr in the MBE chamber before regrowing 100 nm of N++ Ga2O3 at $3x10^{19}$ $cm^{-3}$—reduces the regrowth interface resistance enough to unlock the intrinsic performance of the $\\beta$-(Al,Ga)2O3/Ga2O3 2DEG channel. The evidence is a device with no low-VDS nonlinearity, peak gm of 110 mS/mm at 15 V, 0.5 A/mm on-current, and an f_T·L_G product of 6.1 GHz·μm, described as one of the highest for Ga2O3. The paper also claims that a 100 nm Al2O3 ALD layer passivates surface traps, shown by the absence of current collapse in 200 ns pulsed I-V under gate and dual bias stress, and that RF gain only appears after this passivation. The authors state the f_MAX of 65 GHz is one of the highest reported for Ga2O3-based RF devices.","pith_inferences":["If interface resistance reduction is the root cause, the same pre-regrowth treatment may transfer to other oxide semiconductors where regrown contacts limit RF performance.","The absence of current collapse with a 100 nm Al2O3 layer suggests testing a thickness series could separate surface trap passivation from mechanical stress effects.","The f_T·L_G product still falls short of GaN HEMTs (typically above 10 GHz·μm), so further gains may require channel mobility or saturation-velocity engineering rather than gate scaling alone.","The paper does not report breakdown voltage for the passivated device; combining this regrowth process with field-plate design may yield a device that is simultaneously high-speed and high-voltage."],"forward_implications":["The reported f_T·L_G product of 6.1 GHz·μm sits among the highest for Ga2O3 FETs and implies a saturation velocity near 3×10^6 cm/s.","Al2O3 passivation can eliminate current collapse in beta-Ga2O3 devices, enabling pulsed operation at high drain bias without DC-RF dispersion.","The combination of low-power etch, HCl clean, and UHV anneal can be adopted by other Ga2O3 regrowth processes to reduce contact resistance and improve RF figures.","RF gain is only observed after passivation, indicating that surface traps, not the channel itself, were the dominant limit in prior unpassivated devices."],"supporting_citations":[{"why":"The earlier generation HFET whose high-power etch caused high regrowth interface resistance; the baseline this work improves upon.","marker":"[16]"},{"why":"Shows that MBE vacuum annealing can recover some plasma-etch damage; motivates the annealing step.","marker":"[20]"},{"why":"Reports low-power SF6/Ar plasma etching with low damage and low contact resistance; supplies the etch recipe.","marker":"[21]"},{"why":"First demonstration of the beta-(Al,Ga)2O3/Ga2O3 heterostructure 2DEG channel used here.","marker":"[10]"},{"why":"Provides the open-pad de-embedding method used for RF measurements and extraction of fT/fMAX.","marker":"[28]"},{"why":"Prior Ga2O3-on-SiC RF MOSFET with fT/fMAX of 27.6/57 GHz, used as a benchmark for the reported values.","marker":"[13]"},{"why":"Reports the 71 GHz fMAX record for Ga2O3, used as comparison for the 65 GHz result.","marker":"[14]"},{"why":"Same group's thin-channel Ga2O3 MOSFET with 55 GHz fMAX, providing a prior reference point for fMAX improvement.","marker":"[17]"}],"fun_headline_variants":["No collapse: Ga2O3 HFET hits 32/65 GHz","Vacuum anneal unlocks 65 GHz in Ga2O3 transistor","Ga2O3 HFET hits 32/65 GHz with vacuum anneal","New regrowth process yields 65-GHz Ga2O3 HFET","Ga2O3 device: 32/65 GHz, traps passivated"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reported cutoff frequencies rest on standard open-pad de-embedding and a 20 dB/dec extrapolation from S-parameter data measured only up to 20 GHz; if parasitic pad effects are not fully removed or the gain rolloff is not exactly -20 dB/dec, the extracted f_T = 32 GHz and f_MAX = 65 GHz could be different.","fun_headline_variants_meta":{"raw":{"variants":["No collapse: Ga2O3 HFET hits 32/65 GHz","Vacuum anneal unlocks 65 GHz in Ga2O3 transistor","Ga2O3 HFET hits 32/65 GHz with vacuum anneal","New regrowth process yields 65-GHz Ga2O3 HFET","Ga2O3 device: 32/65 GHz, traps passivated"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000858,"raw_usage":{"total_tokens":3846,"prompt_tokens":1184,"completion_tokens":2662,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":800,"completion_tokens_details":{"reasoning_tokens":2561}},"tokens_in":800,"tokens_out":2662,"duration_ms":16164,"temperature":1.0,"reasoning_tokens":2561,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T11:22:26.969270+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test is to measure S-parameters up to at least 65 GHz on the passivated device and locate the unity-gain crossings of |h21| and unilateral gain without relying on extrapolation; if the measured rolloff deviates from 20 dB/dec or the crossings are below 32 and 65 GHz, the headline numbers would need revision.","supporting_citations":[{"cited_title":"Zhou , author H","cited_arxiv_id":null,"evidence_quote":"The earlier generation HFET whose high-power etch caused high regrowth interface resistance; the baseline this work improves upon."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows that MBE vacuum annealing can recover some plasma-etch damage; motivates the annealing step."},{"cited_title":"Bhattacharyya , author S","cited_arxiv_id":null,"evidence_quote":"First demonstration of the beta-(Al,Ga)2O3/Ga2O3 heterostructure 2DEG channel used here."},{"cited_title":"Vaidya \\ and\\ author U","cited_arxiv_id":null,"evidence_quote":"Provides the open-pad de-embedding method used for RF measurements and extraction of fT/fMAX."},{"cited_title":"Krishnamoorthy , author Z","cited_arxiv_id":null,"evidence_quote":"Prior Ga2O3-on-SiC RF MOSFET with fT/fMAX of 27.6/57 GHz, used as a benchmark for the reported values."},{"cited_title":"Zhang , author A","cited_arxiv_id":null,"evidence_quote":"Reports the 71 GHz fMAX record for Ga2O3, used as comparison for the 65 GHz result."},{"cited_title":"Zhou , author H","cited_arxiv_id":null,"evidence_quote":"Same group's thin-channel Ga2O3 MOSFET with 55 GHz fMAX, providing a prior reference point for fMAX improvement."}],"review_version":1}