arXiv subjects

Runsheng Wang

Publications and source records attributed to Runsheng Wang.

81 records · Page 5Linked to original sources

arXiv2023

The interface states in gate-all-around transistors (GAAFETs)

The atomic-level structural detail and the quantum effects are becoming crucial to device performance as the emerging advanced transistors, representatively GAAFETs, are scaling down towards sub-3nm nodes. However, a multiscale simulation framework based on atomistic models and ab initio quantum simulation is still absent. Here, we propose such a simulation framework by fulfilling three challenging tasks, i.e., building atomistic all-around interfaces between semiconductor and amorphous gate-oxide, conducting large-scale first-principles calculations on the interface models containing up to 2796 atoms, and finally bridging the state-of-the-art atomic level calculation to commercial TCAD. With this framework, two unnoticed origins of interface states are demonstrated, and their tunability by changing channel size, orientation and geometry is confirmed. The quantitative study of interface states and their effects on device performance explains why the nanosheet channel is preferred in industry. We believe such a bottom-up framework is necessary and promising for the accurate simulation of emerging advanced transistors.

physics.app-ph↗

arXiv2023

HybridNet: Dual-Branch Fusion of Geometrical and Topological Views for VLSI Congestion Prediction

Accurate early congestion prediction can prevent unpleasant surprises at the routing stage, playing a crucial character in assisting designers to iterate faster in VLSI design cycles. In this paper, we introduce a novel strategy to fully incorporate topological and geometrical features of circuits by making several key designs in our network architecture. To be more specific, we construct two individual graphs (geometry-graph, topology-graph) with distinct edge construction schemes according to their unique properties. We then propose a dual-branch network with different encoder layers in each pathway and aggregate representations with a sophisticated fusion strategy. Our network, named HybridNet, not only provides a simple yet effective way to capture the geometric interactions of cells, but also preserves the original topological relationships in the netlist. Experimental results on the ISPD2015 benchmarks show that we achieve an improvement of 10.9% compared to previous methods.

cs.LG↗

arXiv2022

CircuitNet: An Open-Source Dataset for Machine Learning Applications in Electronic Design Automation (EDA)

The electronic design automation (EDA) community has been actively exploring machine learning (ML) for very large-scale integrated computer-aided design (VLSI CAD). Many studies explored learning-based techniques for cross-stage prediction tasks in the design flow to achieve faster design convergence. Although building ML models usually requires a large amount of data, most studies can only generate small internal datasets for validation because of the lack of large public datasets. In this essay, we present the first open-source dataset called CircuitNet for ML tasks in VLSI CAD.

cs.LG↗

arXiv2021

Understanding Hot Carrier Reliability in FinFET Technology from Trap-based Approach

In this paper, the recent advances of our studies on hot carrier degradation (HCD) are presented from trap-based approach. The microscopic speculation of interface trap generation is carried out by time-dependent DFT (TDDFT) simulation in "real-time". Two types of oxide traps contributing to HCD are identified from experiments. Combining the contributions of interface and oxide traps, a unified compact model has been proposed which can accurately predict hot carrier degradation and variation in full Vgs/Vds bias. The trap locations, degradation contributions and temperature dependence are studied in detail. In addition, the mixed mode reliability of HCD-BTI coupling through self-heating and under off-state stress are discussed.

cond-mat.mes-hall↗

arXiv2020

DaSGD: Squeezing SGD Parallelization Performance in Distributed Training Using Delayed Averaging

The state-of-the-art deep learning algorithms rely on distributed training systems to tackle the increasing sizes of models and training data sets. Minibatch stochastic gradient descent (SGD) algorithm requires workers to halt forward/back propagations, to wait for gradients aggregated from all workers, and to receive weight updates before the next batch of tasks. This synchronous execution model exposes the overheads of gradient/weight communication among a large number of workers in a distributed training system. We propose a new SGD algorithm, DaSGD (Local SGD with Delayed Averaging), which parallelizes SGD and forward/back propagations to hide 100% of the communication overhead. By adjusting the gradient update scheme, this algorithm uses hardware resources more efficiently and reduces the reliance on the low-latency and high-throughput inter-connects. The theoretical analysis and the experimental results show its convergence rate O(1/sqrt(K)), the same as SGD. The performance evaluation demonstrates it enables a linear performance scale-up with the cluster size.

cs.DC↗

arXiv2019

An Energy-Efficient Mixed-Signal Parallel Multiply-Accumulate (MAC) Engine Based on Stochastic Computing

Convolutional neural networks (CNN) have achieved excellent performance on various tasks, but deploying CNN to edge is constrained by the high energy consumption of convolution operation. Stochastic computing (SC) is an attractive paradigm which performs arithmetic operations with simple logic gates and low hardware cost. This paper presents an energy-efficient mixed-signal multiply-accumulate (MAC) engine based on SC. A parallel architecture is adopted in this work to solve the latency problem of SC. The simulation results show that the overall energy consumption of our design is 5.03pJ per 26-input MAC operation under 28nm CMOS technology.

eess.SP↗

arXiv2019

Memory System Designed for Multiply-Accumulate (MAC) Engine Based on Stochastic Computing

Convolutional neural network (CNN) achieves excellent performance on fascinating tasks such as image recognition and natural language processing at the cost of high power consumption. Stochastic computing (SC) is an attractive paradigm implemented in low power applications which performs arithmetic operations with simple logic and low hardware cost. However, conventional memory structure designed and optimized for binary computing leads to extra data conversion costs, which significantly decreases the energy efficiency. Therefore, a new memory system designed for SC-based multiply-accumulate (MAC) engine applied in CNN which is compatible with conventional memory system is proposed in this paper. As a result, the overall energy consumption of our new computing structure is 0.91pJ, which is reduced by 82.1% compared with the conventional structure, and the energy efficiency achieves 164.8 TOPS/W.

eess.SP↗

arXiv2019

New-Generation Design-Technology Co-Optimization (DTCO): Machine-Learning Assisted Modeling Framework

In this paper, we propose a machine-learning assisted modeling framework in design-technology co-optimization (DTCO) flow. Neural network (NN) based surrogate model is used as an alternative of compact model of new devices without prior knowledge of device physics to predict device and circuit electrical characteristics. This modeling framework is demonstrated and verified in FinFET with high predicted accuracy in device and circuit level. Details about the data handling and prediction results are discussed. Moreover, same framework is applied to new mechanism device tunnel FET (TFET) to predict device and circuit characteristics. This work provides new modeling method for DTCO flow.

cs.OH↗

arXiv2019

A Parallel Bitstream Generator for Stochastic Computing

Stochastic computing (SC) presents high error tolerance and low hardware cost, and has great potential in applications such as neural networks and image processing. However, the bitstream generator, which converts a binary number to bitstreams, occupies a large area and energy consumption, thus weakening the superiority of SC. In this paper, we propose a novel technique for generating bitstreams in parallel, which needs only one clock for conversion and significantly reduces the hardware cost. Synthesis results demonstrate that the proposed parallel bitstream generator improves 2.5x area and 712x energy consumption.

cs.ET↗