skip to main content


Search for: All records

Creators/Authors contains: "Yu, Shimeng"

Note: When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher. Some full text articles may not yet be available without a charge during the embargo (administrative interval).
What is a DOI Number?

Some links on this page may take you to non-federal websites. Their policies may differ from this site.

  1. Abstract

    Cryogenic‐computing draws attention due to its variety of applications such as cloud‐computing, aerospace electronics, and quantum computing. Low temperature (e.g., 77 K) enables higher switching speed, improved reliability, and suppressed noise. Although cryogenic dynamic random‐access memory is studied, the cryogenic NAND flash is not explored intensively. Herein, a cryogenic storage memory based on the charge‐trap mechanism is reported. By removing the tunneling oxide from the conventional silicon/oxide/nitride/oxide/silicon (SONOS)‐type flash memory (therefore becoming silicon/oxide/nitride/silicon (SONS)), high‐speed and low‐power operation is aimed to be achieved while relieved from poor retention issue thanks to the cryogenic environment. The FinFET‐structured SONS memory device is demonstrated experimentally with gate length of 20–30 nm, which can achieve the retention issue (>10 years) with low voltage (≈6.5 V) and high speed (≈5 µs) operation at 77 K. To have a holistic system‐level evaluation, benchmark simulation of an interface between a host microprocessor and solid‐state‐drive is conducted, considering the refrigerator cooling cost and the heat loss via cables across two temperatures (300 and 77 K). The results show that the SONS‐type cryogenic storage system shows over 81% improvement in both latency and power, compared to the SONOS counterpart located at cryogenics.

     
    more » « less
  2. Abstract We present a novel deep neural network (DNN) training scheme and resistive RAM (RRAM) in-memory computing (IMC) hardware evaluation towards achieving high accuracy against RRAM device/array variations and enhanced robustness against adversarial input attacks. We present improved IMC inference accuracy results evaluated on state-of-the-art DNNs including ResNet-18, AlexNet, and VGG with binary, 2-bit, and 4-bit activation/weight precision for the CIFAR-10 dataset. These DNNs are evaluated with measured noise data obtained from three different RRAM-based IMC prototype chips. Across these various DNNs and IMC chip measurements, we show that our proposed hardware noise-aware DNN training consistently improves DNN inference accuracy for actual IMC hardware, up to 8% accuracy improvement for the CIFAR-10 dataset. We also analyze the impact of our proposed noise injection scheme on the adversarial robustness of ResNet-18 DNNs with 1-bit, 2-bit, and 4-bit activation/weight precision. Our results show up to 6% improvement in the robustness to black-box adversarial input attacks. 
    more » « less