Deep neural networks (DNNs) are currently widely used for many artificial intelligence (AI) applications including computer vision, speech recognition, and robotics. While DNNs deliver state-of-the-art accuracy on many AI tasks, it comes at the cost of high computational complexity. Accordingly, techniques that enable efficient processing of DNNs to improve energy efficiency and throughput without sacrificing application accuracy or increasing hardware cost are critical to the wide deployment of DNNs in AI systems. This article aims to provide a comprehensive tutorial and survey about the recent advances toward the goal of enabling efficient processing of DNNs. Specifically, it will provide an overview of DNNs, discuss various hardware platforms and architectures that support DNNs, and highlight key trends in reducing the computation cost of DNNs either solely via hardware design changes or via joint hardware design and DNN algorithm changes. It will also summarize various development resources that enable researchers and practitioners to quickly get started in this field, and highlight important benchmarking metrics and design considerations that should be used for evaluating the rapidly growing number of DNN hardware designs, optionally including algorithmic codesigns, being proposed in academia and industry. The reader will take away the following concepts from this article: understand the key design considerations for DNNs; be able to evaluate different DNN hardware implementations with benchmarks and comparison metrics; understand the tradeoffs between various hardware architectures and platforms; be able to evaluate the utility of various DNN design techniques for efficient processing; and understand recent implementation trends and opportunities.

Efficient Processing of Deep Neural Networks: A Tutorial and Survey

Ultrathin molybdenum disulphide (MoS2) has emerged as an interesting layered semiconductor because of its finite energy bandgap and the absence of dangling bonds. However, metals deposited on the semiconducting 2H phase usually form high-resistance (0.7 kΩ μm–10 kΩ μm) contacts, leading to Schottky-limited transport. In this study, we demonstrate that the metallic 1T phase of MoS2 can be locally induced on semiconducting 2H phase nanosheets, thus decreasing contact resistances to 200–300 Ω μm at zero gate bias. Field-effect transistors (FETs) with 1T phase electrodes fabricated and tested in air exhibit mobility values of ~50 cm2 V−1 s−1, subthreshold swing values below 100 mV per decade, on/off ratios of >107, drive currents approaching ~100 μA μm−1, and excellent current saturation. The deposition of different metals has limited influence on the FET performance, suggesting that the 1T/2H interface controls carrier injection into the channel. An increased reproducibility of the electrical characteristics is also obtained with our strategy based on phase engineering of MoS2. Non-optimal electrical contacts can significantly limit the performance of MoS2-based thin-film transistors. Transformation of semiconducting MoS2 into its metallic phase is now shown as a viable strategy to decrease the metal–MoS2 contact resistance.

Phase-engineered low-resistance contacts for ultrathin MoS2 transistors.

The memory subsystem accounts for a significant cost and power budget of a computer system. Current DRAM-based main memory systems are starting to hit the power and cost limit. An alternative memory technology that uses resistance contrast in phase-change materials is being actively investigated in the circuits community. Phase Change Memory (PCM) devices offer more density relative to DRAM, and can help increase main memory capacity of future systems while remaining within the cost and power constraints.In this paper, we analyze a PCM-based hybrid main memory system using an architecture level model of PCM.We explore the trade-offs for a main memory system consisting of PCMstorage coupled with a small DRAM buffer. Such an architecture has the latency benefits of DRAM and the capacity benefits of PCM. Our evaluations for a baseline system of 16-cores with 8GB DRAM show that, on average, PCM can reduce page faults by 5X and provide a speedup of 3X. As PCM is projected to have limited write endurance, we also propose simple organizational and management solutions of the hybrid memory that reduces the write traffic to PCM, boosting its lifetime from 3 years to 9.7 years.

http://people.cs.uchicago.edu/~ftchong/290N-W10/pcm.pdf

Scalable high performance main memory system using phase-change memory technology

In this paper, we propose and validate a novel design for a double-gate tunnel field-effect transistor (DG tunnel FET), for which the simulations show significant improvements compared with single-gate devices using a gate dielectric. For the first time, DG tunnel FET devices, which are using a high-gate dielectric, are explored using realistic design parameters, showing an on-current as high as 0.23 mA for a gate voltage of 1.8 V, an off-current of less than 1 fA (neglecting gate leakage), an improved average subthreshold swing of 57 mV/dec, and a minimum point slope of 11 mV/dec. The 2D nature of tunnel FET current flow is studied, demonstrating that the current is not confined to a channel at the gate-dielectric surface. When varying temperature, tunnel FETs with a high-kappa gate dielectric have a smaller threshold voltage shift than those using SiO2, while the subthreshold slope for fixed values of Vg remains nearly unchanged, in contrast with the traditional MOSFET. Moreover, an Ion/Ioff ratio of more than 2 times 1011 is shown for simulated devices with a gate length (over the intrinsic region) of 50 nm, which indicates that the tunnel FET is a promising candidate to achieve better-than-ITRS low-standby-power switch performance.

https://www.silvaco.co.kr/tech_lib_TCAD/simulationstandard/2008/jan_feb_mar/a2/jan-feb-mar08_a2.pdf

Double-Gate Tunnel FET With High- $\kappa$ Gate Dielectric

Substantial effort has been placed on developing semiconducting carbon nanotubes and nanowires as building blocks for electronic devices--such as field-effect transistors--that could replace conventional silicon transistors in hybrid electronics or lead to stand-alone nanosystems. Attaching electric contacts to individual devices is a first step towards integration, and this step has been addressed using lithographically defined metal electrodes. Yet, these metal contacts define a size scale that is much larger than the nanometre-scale building blocks, thus limiting many potential advantages. Here we report an integrated contact and interconnection solution that overcomes this size constraint through selective transformation of silicon nanowires into metallic nickel silicide (NiSi) nanowires. Electrical measurements show that the single crystal nickel silicide nanowires have ideal resistivities of about 10 microOmega cm and remarkably high failure-current densities, >10(8) A cm(-2). In addition, we demonstrate the fabrication of nickel silicide/silicon (NiSi/Si) nanowire heterostructures with atomically sharp metal-semiconductor interfaces. We produce field-effect transistors based on those heterostructures in which the source-drain contacts are defined by the metallic NiSi nanowire regions. Our approach is fully compatible with conventional planar silicon electronics and extendable to the 10-nm scale using a crossed-nanowire architecture.

飯田裕幸

Papers

International Technology Roadmap for Semiconductors 2003の要求清浄度について－シリコンウエハ表面と雰囲気環境に要求される清浄度, 分析方法の現状について－

飯田 裕幸

Papers

International Technology Roadmap for Semiconductors 2003の要求清浄度について － シリコンウエハ表面と雰囲気環境に要求される清浄度, 分析方法の現状について －

飯田裕幸

International Technology Roadmap for Semiconductors 2003の要求清浄度について－シリコンウエハ表面と雰囲気環境に要求される清浄度, 分析方法の現状について－