1 September 2025 in Fundamental Research
Advancements in the semiconductor industry introduce novel channel materials, device structures, and integration methods, leading to intricate physics challenges when characterizing devices at circuit level. Nevertheless, accurate models for emerging devices are crucial for physics-driven TCAD-to-SPICE flows to enable the increasingly vital design technology co-optimization (DTCO). Particularly for ultra-scaled devices where quantum effects become significant, this led to the introduction of empirical model parameters and a disconnection to manufacturing processes. To catch up with these developments, an alternative to the traditional white-box modeling methods has attracted much attention: machine learning-assisted compact modeling (MLCM). These black-box methods target towards general-purpose modeling of complex mathematics and physics through training of neural networks on experimental and simulated data, generating an accurate closed-form mapping between output characteristics and input parameters for fabrication process and device operation. To address this new trend, this work provides a comprehensive overview of emerging device model methodologies, spanning from device physics to machine learning engines. By analyzing, structuring, and extending distributed efforts on this topic, it is shown how MLCM can overcome limitations of traditional compact modeling and contribute to effective DTCO to further advance semiconductor technologies.
24 May 2022 in IEEE Transactions on Electron Devices
Complex nonlinear dependence of ultra-scaled transistor performance on its channel geometry and source/drain (S/D) doping profile bring obstacles in the advanced technology path-finding and optimization. A machine learning-based multi-objective optimization (MOO) workflow is proposed to optimize the sub-3-nm node gate-all-around (GAA) three-layer-stacked nanosheet transistors (NSFETs) accounting for the key performance knob of channel geometry and S/D doping profile. The artificial neural network (ANN) is trained to learn the compact current–voltage ( ${I}$ – ${V}$ ) relationship of NSFETs from 3-D technology computer-aided design (TCAD) simulation results. Based on the artificial neural network (ANN) model, MOO between threshold swing, on–off ratio, and ON-state current of NSFETs is performed with adaptive weighted sum theory. The proposed workflow efficiently suggests an optimized design window of channel geometry and doping profile of NSFETs. The proposed devices satisfy the 2025 International Roadmap for Devices and Systems (IRDSs) target in terms of electrical characteristics for digital circuits.
6 June 2022 in IEEE Transactions on Electron Devices
Complementary FET (CFET) is a promising candidate for CMOS scaling beyond 3-nm technology node. In this article, a novel hybrid channel CFET (HC-CFET) is proposed, which takes advantage of the vertical structure and simultaneously co-optimizes the preferred high-electron/hole-mobility surface of NMOS/PMOS on one substrate. The flexible combination of nanowires (NWs) and nanosheets (NSs) in the HC-CFET allows NMOS to have (100) channel surface orientation, while PMOS has (110) channel surface orientation without increasing the footprint of CFET pillars. Parasitic-aware device to circuit design-technology co-optimization (DTCO) analysis of the HC-CFET is performed based on advanced TCAD simulation of the device and comprehensive HSPICE simulation with TCAD-calibrated compact model. At the device level, the optimization of the channel surface orientation in the HC-CFET enables both NMOS and PMOS to obtain a current gain of more than 20%. By adjusting the wafer orientation and stacking type, HC-CFET exhibits higher frequency gain and comparable energy consumption in ring oscillator (RO) than multiscreen CFET (MS-CFET) and multibridge CFET (MB-CFET). After fully considering the impact of the stacking type on the static random access memory (SRAM) cell structure, (110) wafer together with n-NW on top of p-NS stack is the preferable approach to manufacture HC-CFET, which helps balance the speed, stability, and area of the SRAM cell.
30 September 2022 in IEEE Transactions on Electron Devices
Complementary FET (CFET) is a promising booster for further area reductions in static random-access memory (SRAM) cells. However, the performance degrading by a series of parasitic parameters in these SRAM cells will diminish the scaling benefit introduced by new transistor architectures. In this article, we propose a universal methodology to determine the layout optimization direction of 6T static random-access memory (6T-SRAM) cells by studying the sensitivity of 6T-SRAM cell performance to various parasitic parameters. And adopt this method to optimize the CFET SRAM cell layout structure under the advanced nodes beyond 3 nm. The performances of CFET 6T-SRAM cells with different layout schemes are comparatively evaluated. It is found that the influence of the parasitic resistance on 6T-SRAM performances is double-sided, and parasitic resistances from transistor to power rail ( ${R}_{\mathrm {pds}}$ , ${R}_{\mathrm {pud}}$ ) and from access (AX) devices to bit-lines (BLs) ( ${R}_{\mathrm{bax}}$ ) have the most important effect on improving read noise margin (RNM) together with write margin (WM). The BL-first scheme with reduced ${R}_{\mathrm{bax}}$ and increased ${R}_{\mathrm {pud}}$ is proven to have 76.3% improvement in WM and 122.5% decrease in write time (WT) compared to the BL-last scheme, and a 63.8% improvement in RNM compared to conventional nanosheet architecture. Further optimized ${R}_{\mathrm{bax}}$ in the buried-BL scheme is proven to have higher WM, as well as lower WT. The BL-first scheme and buried-BL scheme is proven to be the most efficient approach to boost CFET SRAM performance.
23 September 2022 in IEEE Transactions on Electron Devices
In this article, one feasible fabrication approach for novel fishbone FETs using the channel-first and single work function metal (sWFM) processes is proposed and investigated by 3-D technical computer-aided design (TCAD) simulations. Through a small modification on the fabrication process of general gate-all-around (GAA) nanosheet FETs (NSFETs), the special fishbone-like channel composed of vertically stacked Si NSs and sandwiched SiGe nano-fins is experimentally demonstrated by the channel-first process. The simulated electrical characteristics show that the width of the nano-fins should be within 5 nm for a better gate control. Unlike traditional NSFETs, symmetrical threshold voltages ( ${V}_{\text {th}}\text{s}$ ) for n-type and p-type fishbone FETs can be achieved by using a sWFM, and $\Delta {V}_{\text {th}}$ is optimized by 99.26% compared with that of NSFETs. Meanwhile, it is also found that the SiGe nano-fins contribute more driving current for p-type devices. Therefore, the proposed fishbone FETs with sWFM not only exhibit significantly enhanced driving current but also provide good balance between the performances of n-type and p-type fishbone FETs with a little extra process cost.
23 October 2023 in IEEE Transactions on Electron Devices
Complementary field-effect transistor is considered to be one of the most promising structures to replace gate-all-around (GAA) field-effect transistors after 2 nm node and extend Moore’s Law. In this work, we develop the nanosheet (NS) count design strategy to address the mobility mismatch of electron and hole in standard complementary FET (CFET) cells, that is, nine-stage ring oscillator (RO), 6T-SRAM cell. The device characteristics are obtained by state-of-the-art physics-based TCAD simulations, and high-accuracy 3-D parasitic extraction is implemented in the device to circuit co-optimization flow. The simulation result of RO and SRAM circuits reveals that altering the ratio of nFET NSs (nNSs) and pFET NSs (pNSs) can match the mobility of electrons and holes, thus optimizing the performance and power of different circuits for different applications. For CFETs in nine-state RO cells, the CFET with four pNS channels and two nNS channels shows a 10% increase in frequency and 6% decrease in capacitance, but only a 3% increase in power consumption when compared to the CFET with three pNS channels and three nNS channels. Also take the “3 $+$ 3” CFET as benchmark, for CFETs in SRAM cells, the “4 $+$ 2” CFET delivers 10% extra static noise margin (SNM) and 8% $\textit{I}_{\text{leak}}$ decrease, but the “1 $+$ 2” CFET offers 12% extra SNM with 20% $\textit{I}_{\text{leak}}$ increase, and advantage in terms of manufacturing process complexity.