betaFeedback

WorksPhysics and Astronomy

Agentic TCAD Calibration Workflow for Oxide Semiconductor Transistors

Abstract

Experimental TCAD calibration is essential for predictive technology modeling of emerging oxide semiconductor transistors. However, it remains time-consuming and expert dependent because of model ambiguity. Multiple physical models and parameter sets can reproduce the same measured transfer characteristics, while local fitting alone cannot uniquely identify the underlying device physics. We present the first demonstration of an agentic TCAD calibration workflow for a fabricated bottom-gate In–W–O (BG-IWO) transistor. Starting from the measured transfer curve and device information, the workflow uses measurement–TCAD residuals and local sensitivity tests to select bounded parameter corrections or evaluate additional physical models, and accept only updates that improve device metrics. The LLM agent orchestrates the workflow, while Sentaurus governs the device physics. For the 2%-W reference device, five agent-suggested updates yield a fixed calibrated model, reducing the multi-metric device objective JJ by 14.3×\times. Maximum VthV_{\mathrm{th}}/IonI_{\mathrm{on}} errors are 36.1~mV/0.022 decade for varying-drain-bias tests and 46.2~mV/0.062 decade for varying-channel-length tests, demonstrating model transferability across bias and geometry rather than a local parameter fit. W-composition tests provide process-sensitive insight. This agentic workflow provides a faster route to model development for emerging device technologies.

Introduction

Oxide semiconductor transistors have emerged as BEOL-compatible, low-thermal-budget candidates for CMOS+X applications such as memory access transistors and power converters [1, 2]. For emerging devices under short-loop experimentation, turning measured transfer characteristics into a predictive TCAD deck for future process optimization remains a nontrivial inverse device-modeling problem. In practice, calibration becomes a time-consuming and expert-dependent bottleneck because each iteration requires physical-model selection, parameter adjustment, deck implementation, Sentaurus execution, and interpretation of measurement–simulation residuals. The measured transfer characteristics reflect coupled electrostatic, transport, and trapping effects, while their TCAD reproduction also depends on device geometry, parameter values, meshing, and numerical implementation.

Gyujun Jeong, Junmo Lee, Sungwon Cho, Asif Khan, and Shimeng Yu are with the School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA (e-mail: [email protected]; [email protected]).

Woohyun Hwang, Kwangyou Seo, Suhwan Lim, Wanki Kim, and Daewon Ha are with Semiconductor Research and Development, Samsung Electronics Co., Ltd., South Korea.

Rishi Ranade, Kihang Youn, Ram Cherukuri, and Yiyi Wang are with NVIDIA, Santa Clara, CA, USA.

Figure 1 frames this bottleneck as TCAD model-search ambiguity. Multiple physics models and parameter choices can fit the same measured transfer curve, while a local parameter fit alone does not identify the underlying device physics or guarantee model generalization. This work therefore reduces the active search effort by using measurement–TCAD residuals to rank TCAD physical models and parameter groups, run bounded Sentaurus tests, and accept only updates that improve predefined device metrics.

Motivation for agentic TCAD calibration

Figure 1. Motivation for agentic TCAD calibration. Manual calibration is a nontrivial inverse problem because multiple TCAD physics models can fit the same measured transfer curve. This work uses measurement–TCAD residuals to rank model classes, run Sentaurus, and accept only updates that improve predefined device metrics, orchestrated by an AI agent. The workflow produces a fixed calibrated model, reduces model/parameter ambiguity, and provides physical and process insight for follow-up TCAD studies.

Prior machine learning (ML) and artificial intelligence (AI)-assisted TCAD and modeling studies have addressed adjacent tasks that differ from this work, including TCAD–ML variation and temperature analysis [3], inverse parameter calibration [4], and large language model (LLM)-based agentic TCAD code generation and device optimization [5]. However, these earlier works operate within predefined physics, parameter, or design spaces. Hence, they do not provide physical insights for automating model selection and parameter calibration based on experimental data. Table I summarizes the distinctions from prior AI/ML-assisted TCAD studies.

To the best of our knowledge, this work presents the first experimental demonstration of an agentic TCAD calibration workflow. The workflow integrates (i) physics-model selection using retrieval-augmented generation (RAG) [6], (ii) on-the-fly TCAD execution and evaluation of intermediate results, (iii) residual-guided selection of subsequent bounded calibration tests, and (iv) extraction of device-physics and process insights, orchestrated by the AI agent.

AspectTCAD–ML [3]Inverse TCAD calibration [4]AgenticTCAD [5]This work
ObjectiveAnalyze device variation and temperature effectsInfer parameters of a predefined TCAD modelGenerate TCAD code and optimize device designsExperimentally calibrate TCAD physics and parameters to measured II–VV
InputsTCAD-generated II–VV databaseTCAD-generated training data and measured TFT II–VVNatural-language design specificationsMeasured II–VV, device information, and TCAD residuals
Search spaceFixed TCAD physics, sampled variation spacePredefined physics model and fitted parametersPredefined design spaceReduced search space by active physics model selection by reported physical insights
Decision basisML surrogate predictionAI-based inverse fittingLLM agent–simulator iterationAgentic workflow with residual-guided parameter and model selection
ValidationExperimental demonstrationTFT calibration casesSimulation-defined target specificationsDrain-bias and channel-length transfer tests without retuning

Table 1. Comparison with prior AI/ML-assisted TCAD studies.

Agentic TCAD Calibration Workflow

Workflow Overview and Responsibilities

Figure 2(a) shows the residual-guided TCAD calibration loop. The workflow compares the measured transfer curve with the current TCAD result, extracts the residuals and device metrics, and determines whether the next action should adjust an existing parameter or evaluate an additional physics model. RAG retrieves supporting evidence from a knowledge base (KB) comprising the Sentaurus Device User Guide [7] and reviewed device literature. The proposed parameter or physics-model update is translated into Sentaurus syntax, executed, and evaluated using the multi-metric device objective JJ defined in II-C. Each update is accepted only when it reduces JJ. Otherwise, the update is rejected, and the workflow recommends the next bounded TCAD test.

Residual-guided TCAD calibration workflow

Figure 2. Residual-guided TCAD calibration loop. (a) The agentic workflow demonstrates how residuals guide the next TCAD update. The agent compares the measured transfer curve with the current TCAD model, extracts residuals, and suggests either a parameter adjustment or model activation. (b) The example shows how a subthreshold swing residual (Δ\DeltaSS) drives Gaussian/tail DOS activation, improving the multi-metric device objective JJ.

Figure 2(b) illustrates an example flow. The remaining mismatch is mainly in the subthreshold swing, as indicated by Δ\DeltaSS. A RAG search of the Sentaurus manual and device literature identifies Gaussian and tail DOS models whose documented effects are consistent with the measured subthreshold shape. After human review and checks of the parameter ranges and SDevice syntax, these terms are added to the deck. The subsequent Sentaurus run reduces JJ. In the calibration loop, bounded one-factor-at-a-time (OFAT) probes identify existing parameter groups that affect the dominant residual. The selected residual vector then guides derivative-free Gauss–Newton (DFO-GN) updates [8], which are suitable for expensive live-TCAD residual evaluations. Section II-C defines the local sensitivity, and the DFO-GN update. Section II-D describes when an additional physical model is considered and how each update is verified.

Table II distinguishes the responsibilities of the LLM agent, Sentaurus TCAD, and the human reviewer. In an LLM-agent workflow, the language model can coordinate external tools and use the returned results to select subsequent actions [9]. Here, the LLM agent serves as the workflow orchestrator rather than the device solver. Sentaurus TCAD solves the device equations and computes the device response for each proposed update. The human reviewer defines and approves the TCAD workflow instructions provided to the agent through Agent Skills [10], including permitted deck modifications, calibration-metric checks, and numerical-recovery procedures. The workflow logic is implemented independently of the selected LLM backend. The reported implementations used OpenAI GPT-5.5 [11] as the backend LLM and executed the workflow through OpenAI Codex [12] and the open-source

ComponentResponsibility
LLM agentAnalyze residuals and device metrics based on the retrieved RAG/KB evidence, and propose a bounded TCAD update.
Sentaurus TCADExecute the human-approved deck and return the simulated transfer characteristics and convergence status.
Human reviewerProvide the device stack and test splits, approve agent-proposed updates, and verify their physical validity and interpretation.

Table II. Responsibilities in the agentic TCAD workflow.

NVIDIA NeMo Agent Toolkit [13]. All device simulations were performed using Synopsys Sentaurus W-2024.09-SP1.

Device Structure and Experimental Design

The experimental target is a fabricated bottom-gate In–W–O (BG-IWO) transistor measured using forward IDI_D–VGV_G sweeps. Table III summarizes the device dimensions, materials, and measurement configuration. Figure 3(a) shows the fixed 2D TCAD geometry and its SVisual visualization, and Figure 3(b) provides SEM dimension references. Device geometry and basic material information are fixed before fitting, while the underlying physical model parameters are to be calibrated.

Fixed 2D TCAD geometry schematic and SEM top view of a BG-IWO transistor

Figure 3. (a) Fixed 2D TCAD geometry schematic and SVisual visualization of the BG-IWO. (b) SEM top view of an L=2L = 2 µm BG-IWO transistor used as a dimension reference.

ParameterSpecification
MeasurementForward IDI_D–VGV_G sweep
Channel materialAmorphous In–W–O (IWO): 2% W (calibration) and 4% W (process test)
Channel length, LL1.0 µm (calibration) and 2.0 µm (length test)
Channel width, WW5.0 µm
S/D overlap, LovL_{\mathrm{ov}}1.5 µm per side (3.0 µm total)
IWO thickness3.5 nm
Gate dielectricHfO₂, 7.0 nm
PassivationConformal HfO₂, 5.0 nm
S/D contactsPd, 40 nm

Table III. Fabricated BG-IWO transistor and device specifications used for calibration.

As specified in Table IV, nominal calibration uses the two measured IDI_D–VGV_G transfer curves acquired at VD∈BcalV_D \in\mathcal{B}_{\mathrm{cal}}, where Bcal={50 mV,1 V}\mathcal{B}_{\mathrm{cal}} = \{50\ \mathrm{mV}, 1\ \mathrm{V}\}. Figure 4 shows the metric windows used to evaluate these errors and extract VthV_{\mathrm{th}}, IonI_{\mathrm{on}}, and SS. The same current floor, gate-voltage windows, and metric-extraction method are applied at every parameter setting. Only the nominal 2%-W, L=1 μmL = 1\ \mu\mathrm{m} curves measured at biases b∈Bcalb \in\mathcal{B}_{\mathrm{cal}} contribute to JJ. All other drain-bias and channel-length conditions are reserved for evaluating the final calibrated model without retuning.

SplitConditions and model use
Nominal calibration2% W, L=1.0 μmL = 1.0\ \mu\mathrm{m}, VD∈BcalV_D \in\mathcal{B}_{\mathrm{cal}}, where Bcal={0.05,1.00} V\mathcal{B}_{\mathrm{cal}} = \{0.05, 1.00\}\ \mathrm{V}. Used for nominal calibration objective JJ
Drain-bias transfer2% W, L=1.0 μmL = 1.0\ \mu\mathrm{m}, VD={0.05,0.10,0.20,0.50,1.00} VV_D = \{0.05, 0.10, 0.20, 0.50, 1.00\}\ \mathrm{V} for varying-bias-test without retuning
Channel-length transfer2% W, L=2.0 μmL = 2.0\ \mu\mathrm{m}, VD={0.05,1.00} VV_D = \{0.05, 1.00\}\ \mathrm{V} for varying-channel-length test without retuning
W-composition test4% W, L=1.0 μmL = 1.0\ \mu\mathrm{m}, VD={0.05,1.00} VV_D = \{0.05, 1.00\}\ \mathrm{V}. Separately recalibrated for process sensitivity test

Table IV. Measurement data split.

Calibration metric windows for objective J

Figure 4. Calibration metric windows for objective JJ: log/linear-current, VthV_{\mathrm{th}}, IonI_{\mathrm{on}}, and SS residuals define the calibration objective in Eq. (1).

Calibration objective JJ and sensitivity-based parameter adjustment

For the drain-bias set Bcal\mathcal{B}_{\mathrm{cal}} defined in Table IV, the multi-metric calibration objective for a TCAD model MM with parameter vector θ\boldsymbol{\theta} is

J(M,θ)=1∣Bcal∣∑b∈Bcal[wlogElog,b+wlinElin,b+wth∣ΔVth,b∣+won∣Δlog⁡10Ion,b∣+wSS∣ΔSSb∣].(1)J(M,\boldsymbol{\theta}) = \frac{1}{\lvert\mathcal{B}_{\mathrm{cal}}\rvert} \sum_{b \in\mathcal{B}_{\mathrm{cal}}} \left[ w_{\mathrm{log}}E_{\mathrm{log},b} + w_{\mathrm{lin}}E_{\mathrm{lin},b} + w_{\mathrm{th}}\lvert\Delta V_{\mathrm{th},b}\rvert+ w_{\mathrm{on}}\lvert\Delta\log_{10} I_{\mathrm{on},b}\rvert + w_{\mathrm{SS}}\lvert\Delta\mathrm{SS}_{b}\rvert\right]. \tag*{(1)}

Here, bb denotes the drain bias, and Elog,bE_{\mathrm{log},b} and Elin,bE_{\mathrm{lin},b} are the root-mean-square errors (RMSEs) between the simulated and measured log⁡10∣ID∣\log_{10}|I_D| and ∣ID∣|I_D|, respectively. As illustrated in Figure 4, VthV_{\mathrm{th}} is extracted at ∣ID∣=100|I_D| = 100 nA, SS is obtained from the average inverse slope between 0.1 and 100 nA, and IonI_{\mathrm{on}} is the maximum ∣ID∣|I_D| in the transfer curve. The linear-current error is evaluated within one decade of IonI_{\mathrm{on}}, while currents below the 10 pA measurement floor are excluded from the log-current error. Each Δ\Delta term denotes the simulation–measurement difference in the corresponding extracted metric. The fixed weights in Eq. (1) are wlog⁡=0.25w_{\log} = 0.25, wlin=0.10w_{\mathrm{lin}} = 0.10, wth=1.00w_{\mathrm{th}} = 1.00, won=0.20w_{\mathrm{on}} = 0.20, and wSS=0.25w_{\mathrm{SS}} = 0.25. The same extraction rules are applied to every measured and simulated curve, and each objective term is normalized and weighted using values fixed before calibration.

The normalized signed residual and local sensitivity matrix for parameter group PP are

ri=si(M,θ)−miai,SP=∂r∂xP,(2)r_i = \frac{s_i(M,\theta)-m_i}{a_i}, \qquad S_P = \frac{\partial r}{\partial x_P}, \tag*{(2)}

where sis_i and mim_i are corresponding simulated and measured current values or extracted device metrics, ai>0a_i > 0 is a fixed normalization factor chosen so that rir_i is dimensionless, and xPx_P denotes the normalized TCAD parameter for group PP.

The scalar objective JJ quantifies the overall fit between the simulated and measured transfer characteristics. The magnitude, sign, and gate voltage (VGV_G) dependence of rr distinguish among a VthV_{\mathrm{th}} offset, a drain current (IDI_D) magnitude mismatch, and an SS mismatch. The sensitivity SPS_P indicates which existing TCAD parameter group can correct the dominant residual. Within the approved parameter bounds, OFAT Sentaurus simulations estimate the local sensitivity matrix SPS_P by varying one TCAD parameter at a time and evaluating the resulting finite-difference change in the residual vector.

For a parameter group PP selected by the OFAT sensitivity ranking, DFO-GN calculates the bounded parameter change

sP∗=arg min⁡ℓP≤xP+s≤uP 12∥W1/2(r+SPs)∥22(3)s_P^\ast= \underset{\ell_P \le x_P+s \le u_P}{\operatorname{arg\,min}} \ \frac{1}{2}\left\|W^{1/2}(r+S_Ps)\right\|_2^2 \tag*{(3)}

Here, ℓP\ell_P and uPu_P are the allowed parameter limits, ΔP\Delta_P limits the parameter change in one update, and WW contains fixed residual weights. Only parameters in group PP are changed. The local least-squares problem calculates a trial parameter setting. The proposed change is evaluated by a full Sentaurus run, and the resulting JJ determines whether the fit improves.

Physics-Model Activation and Update Verification

An additional physics model is considered only when a systematic transfer-curve mismatch persists after bounded adjustment of the existing parameters. The RAG search retrieves candidate models from the Sentaurus manual and reviewed device literature whose documented behavior is consistent with the remaining mismatch. The human reviewer verifies the applicability of the new physics model and parameter range before approving any deck modification. A parameter or model update is accepted only when the Sentaurus run converges and reduces JJ. Nonconverged trials undergo a predefined numerical-recovery sequence, such as adjusting the bias-step size, iteration limit, or solver tolerance. If convergence cannot be recovered, the run is recorded as a numerical failure and excluded from model-update decisions.

Experimental Calibration and Model Transfer Test

Residual-Guided Calibration of the Reference Device

Figure 5 summarizes how five accepted updates yield a calibrated BG-IWO TCAD model. All TCAD results in this section use the fixed 2D structure defined in Table III. The agent suggests each tuning step by evaluating the residual components after each Sentaurus run. The baseline has a multi-metric objective of J=2.589J = 2.589.

Agent-suggested residual-to-physics updates and calibrated BG-IWO TCAD model

Figure 5. Agent-suggested residual-to-physics updates yield a calibrated BG-IWO TCAD model. (a) Five accepted updates reduce JJ from 2.589 to 0.181. (b) Stepwise transfer curves show progressive correction. (c) Calibrated electrostatic and DOS panels show how agentic actions address systematic residuals. (d) The update table links each residual component to the agent-suggested TCAD action and model class. Yellow highlights mark model activations, showing that QfQ_f and DOS terms are introduced only after parameter adjustments leave systematic residual errors. (e) The residual heatmap shows which error components are reduced at each calibration step. Each value is normalized to its Step 0 value for visualization (Step 0 = 1).

In Step 1, the agent identifies a low-current residual and suggests lowering the source/drain effective WF from 4.7 to 4.16 eV, which reduces JJ to 1.537. Because the dominant residual in Step 2 is a threshold offset, the agent suggests lowering the gate WF from 4.4 to 4.0 eV and enabling a gate-interface fixed-charge term Qf=+1.59×1010 cm−2Q_f = +1.59 \times10^{10}\,\mathrm{cm}^{-2}, reducing JJ to 0.574. In Step 3, the residual indicates a charge deficit, leading the agent to increase ND,effN_{D,\mathrm{eff}} from 1.0×10181.0 \times10^{18} to 2.56×1018 cm−32.56 \times10^{18}\,\mathrm{cm}^{-3}, reducing JJ to 0.443. The left electrostatic/charge panel in Figure 5(c) shows that the WF, QfQ_f, and ND,effN_{D,\mathrm{eff}} updates first correct the electrostatic/charge response.

After these updates, the remaining error is mainly a SS residual, as shown in Figure 5(b) and Figure 5(e). Step 4 therefore introduces the Gaussian and tail effective DOS descriptors shown in the right DOS panel of Figure 5(c), reducing JJ to 0.282. The Gaussian term has NDOS=2.19×1019 cm−3eV−1N_{\mathrm{DOS}} = 2.19 \times10^{19}\,\mathrm{cm}^{-3}\mathrm{eV}^{-1}, centered at EC−0.12E_C - 0.12 eV, and σ=0.08\sigma= 0.08 eV; the tail term has Ntail=7.8×1019 cm−3eV−1N_{\mathrm{tail}} = 7.8 \times10^{19}\,\mathrm{cm}^{-3}\mathrm{eV}^{-1}, Etail=0.0E_{\mathrm{tail}} = 0.0 eV, and σtail=0.133\sigma_{\mathrm{tail}} = 0.133 eV. In Step 5, the agent suggests a final transport adjustment, changing μ0\mu_0 from 10 to 24 cm2/Vs\mathrm{cm}^2/\mathrm{Vs}, yielding the final calibrated model with J=0.181J = 0.181. The fitted variables are calibration-effective TCAD descriptors—effective work-function (WF) terms, gate-interface fixed charge (QfQ_f), effective donor-like background density (ND,effN_{D,\mathrm{eff}}), Gaussian/tail DOS terms, and mobility prefactor (μ0\mu_0)—not direct extractions of defect density, oxygen-vacancy concentration, microscopic mobility, or trap parameters [14].

Fixed-Model Evaluation Across Bias and Geometry

Figure 6 supports generalization of the 2%-W device model beyond the nominal calibration condition. The same fixed TCAD parameter set is used for the varying-drain-bias tests at VD={0.05,0.10,0.20,0.50,1.00}V_D = \{0.05, 0.10, 0.20, 0.50, 1.00\} V and the varying-channel-length tests at L={1,2} μmL = \{1, 2\}\,\mu\mathrm{m} without retuning as described in Table IV. The average and maximum errors are 24.9 and 36.1 mV, respectively, in VthV_{\mathrm{th}}, and 0.013 and 0.022 decade, respectively, in IonI_{\mathrm{on}} for the drain-bias tests. For the channel-length tests, the corresponding average and maximum errors are 27.4 and 46.2 mV in VthV_{\mathrm{th}} and 0.050 and 0.062 decade in IonI_{\mathrm{on}}. Because these sets are defined before calibration, the results demonstrate transferability across bias and geometry, rather than another local parameter fit.

Fixed TCAD model evaluation across drain bias and geometry

Figure 6. Fixed TCAD model evaluation across drain bias and geometry. (a) Varying-drain-bias and (b) varying-channel-length tests of the fixed calibrated 2%-W BG-IWO model. The same TCAD physics models and parameter set are used without retuning, demonstrating model transfer beyond the nominal calibration conditions.

Process-Sensitive Insight

Figure 7 uses the 2%-W model as the reference and evaluates the separately recalibrated 4%-W split for process-composition sensitivity. Figure 7(a) shows that the 4%-W device has a positive VthV_{\mathrm{th}} shift and lower current relative to the 2%-W reference. The observed W-composition trend is consistent with prior InOx_x studies showing that stronger dopant–oxygen bonding reduces sensitivity to oxygen processing and can suppress carrier-density increases associated with oxygen release [15]. In TCAD, this trend is represented by a reduced effective donor-like background density ND,effN_{\mathrm{D,eff}}, rather than by microscopic extraction of oxygen-vacancy density.

W-composition recalibration test

Figure 7. W-composition recalibration test identifies process-sensitive insight. (a) IDI_D–VGV_G characteristics of BG-IWO transistors with different W compositions: 2% reference vs. 4% variation. The 4%-W device shows a positive VthV_{\mathrm{th}} shift and lower current relative to the 2%-W reference. The 4%-W fitting and metrics use its measured range, VG≥−0.7V_G \ge-0.7 V. (b) The agent-suggested descriptor sequence assigns the dominant VthV_{\mathrm{th}} and current shifts to a lower effective donor-like background density ND,effN_{D,\mathrm{eff}}, the residual threshold offset to gate electrostatics, and the subthreshold shape to broadening of the Gaussian and tail DOS terms. (c) Sequential TCAD progress shows each descriptor moving the curve toward the 4%-W data (VD=1V_D = 1 V).

The agent-suggested sequence in Figure 7(b) proceeded as follows. To address the dominant positive VthV_{\mathrm{th}} shift and current reduction observed in the 4%-W device relative to the 2%-W reference, Step 1 reduced ND,effN_{\mathrm{D,eff}} from 2.56×10182.56 \times10^{18} to 6.39×1016 cm−36.39 \times10^{16}\ \mathrm{cm}^{-3}. To correct the residual VthV_{\mathrm{th}} offset through gate electrostatics, Step 2 increased the gate WF from 4.00 to 4.45 eV and changed QfQ_f from +1.59×1010+1.59 \times10^{10} to −1.00×1012 cm−2-1.00 \times10^{12}\ \mathrm{cm}^{-2}. To reduce the remaining SS residual, Step 3 broadened the acceptor-tail distribution from σtail=0.133\sigma_{\mathrm{tail}} = 0.133 to 0.238 eV, increased the donor-like Gaussian DOS from 2.19×10192.19 \times10^{19} to 4.40×1019 cm−3eV−14.40 \times10^{19}\ \mathrm{cm}^{-3}\mathrm{eV}^{-1}, shifted its center toward ECE_C from EC−0.122E_C - 0.122 to EC−0.080E_C - 0.080 eV, and increased its width from σ=0.080\sigma= 0.080 to 0.224 eV.

Figure 7(c) shows the sequential TCAD IDI_D–VGV_G progress as these descriptors move the curve toward the 4%-W data. Here, the calibrated TCAD descriptor updates provide a model-level inverse diagnostic of W-composition sensitivity by identifying the model and parameter classes that reduce distinct components of the 4%-W measurement–TCAD residual. The resulting sequence provides process-sensitive insight that can guide subsequent process experiments and TCAD studies.

Conclusion

An agentic TCAD calibration workflow is demonstrated for fabricated BG-IWO oxide semiconductor transistors using measured IDI_D–VGV_G. The framework converts device context into a human-reviewed TCAD calibration setup and constrains deck updates through reviewed model classes and parameters. For the reference 2%-W model, the workflow reduces the device objective JJ by 14.3× through the residual-guided calibration loop. The fixed calibrated model passes drain-bias and channel-length transferability tests. The 4%-W process-composition test provides process-sensitive insight by ranking the TCAD model classes that reduce distinct residual components associated with the W-composition change. This agentic TCAD calibration framework uses an LLM agent that orchestrates the workflow, while validity is established by the reviewed model classes, live Sentaurus execution, and transferability tests. The proposed methodology is extendable to other emerging device technology pathfinding to allow a fast iterative loop between model generation and process optimization.

Acknowledgment

This work was sponsored by Samsung Electronics (IO250304-12193-01) and by in-kind support from NVIDIA. AI-use disclosure: The authors used generative AI to assist the agentic TCAD workflow study. All the reported results were reviewed and verified by the authors.

References

  1. [1]H. Ye et al., “Double-Gate W-Doped Amorphous Indium Oxide Transistors for Monolithic 3D Capacitorless Gain Cell edRAM”, in IEDM Tech. Dig., Dec. 2020, pp. 28.3.1–28.3.4, doi: 10.1109/IEDM13553.2020.9371981.DOI
  2. [2]S. Deng et al., “BEOL Compatible Oxide Power Transistors for On-Chip Voltage Conversion in Heterogenous 3D (3D) Integrated Circuits,” in IEDM Tech. Dig., Dec. 2023, pp. 1–4, doi: 10.1109/IEDM45741.2023.10413838.DOI
  3. [3]H.-Y. Wong et al., “TCAD-Machine Learning Framework for Device Variation and Operating Temperature Analysis With Experimental Demonstration,” IEEE J. Electron Devices Soc., vol. 8, pp. 992–1000, 2020, doi: 10.1109/JEDS.2020.3024669.DOI
  4. [4]M. H. Ng et al., “AI-Based General Solution for Automated TCAD Parameter Calibration: Case Studies on TFT Devices,” IEEE Trans. Electron Devices, vol. 72, no. 12, pp. 6757–6764, Dec. 2025, doi: 10.1109/TED.2025.3616083.
  5. [5]G. Fan, T. Ma, X. Sun, X. Wang, K. L. Low, and L. Shao, “AgenticTCAD: A LLM-Based Multi-Agent Framework for Automated TCAD Code Generation and Device Optimization,” in Proc. 2026 Design, Automat. Test Eur. Conf. (DATE), Apr. 2026, pp. 1–7, doi: 10.23919/DATE69613.2026.11539536.DOI
  6. [6]P. Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” in Adv. Neural Inf. Process. Syst., vol. 33, 2020, pp. 9459–9474.arxiv.org/abs/2005.11401
  7. [7]Synopsys, Inc., Sentaurus Device User Guide, ver. W-2024.09, Sep. 2024.
  8. [8]C. Cartis and L. Roberts, “A derivative-free Gauss–Newton method,” Math. Program. Comput., vol. 11, no. 4, pp. 631–674, Dec. 2019, doi: 10.1007/s12532-019-00161-7.DOI
  9. [9]L. Wang et al., “A survey on large language model based autonomous agents,” Front. Comput. Sci., vol. 18, Art. no. 186345, 2024, doi:10.1007/s11704-024-40231-1.DOI
  10. [10]NVIDIA, “NVIDIA-Verified Agent Skills,” NVIDIA Skill Documentation. Accessed: Jul. 2, 2026. [Online]. Available: https://docs.nvidia.com/skills
  11. [11]OpenAI, “GPT-5.5 System Card,” Apr. 23, 2026. Accessed: Aug. 15, 2026. [Online]. Available: https://openai.com/index/gpt-5-5-system-card/
  12. [12]N. Bonamy and D. Choi, “Codex as a platform: Build on the open agent harness,” OpenAI Developers, Aug. 19, 2026. Accessed: Sep. 3, 2026. [Online]. Available: https://developers.openai.com/blog/codex-as-a-platform
  13. [13]NVIDIA, NVIDIA NeMo Agent Toolkit, ver. 1.8.0, GitHub repository, Jun. 16, 2026. Accessed: Aug. 28, 2026. [Online]. Available: https://github.com/NVIDIA/NeMo-Agent-Toolkit/releases/tag/v1.8.0
  14. [14]T. Kamiya, K. Nomura, and H. Hosono, “Present status of amorphous In–Ga–Zn–O thin-film transistors,” Sci. Technol. Adv. Mater., vol. 11, no. 4, Sep. 2010, Art. no. 044305, doi: 10.1088/1468-6996/11/4/044305.DOI
  15. [15]S. Aikawa, T. Nabatame, and K. Tsukagoshi, “Effects of dopants in InOx-based amorphous oxide semiconductors for thin-film transistor applications,” Appl. Phys. Lett., vol. 103, no. 17, Oct. 2013, Art. no. 172105, doi: 10.1063/1.4822175.DOI

Paper details

vlnc:W_hr6xhde9
Version
v2 · latest
Published
Added
License
CC BY 4.0
Usage

views · 0

downloads · 4

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

generative AI

View statement
Looking for related papers…
Contents