24 August 2020 in IEEE Transactions on Electron Devices
The requirement of multilevel cell (MLC) resistive random access memory (RRAM) for computing is different than that for MLC storage. It generally requires a linearly spaced conductance median and an ultratight conductance distribution, as the column current are summed up for analog computation. In this article, 3-bit per cell RRAM that is suitable for accurate inference of a deep neural network (DNN) is demonstrated, with ultratight conductance distribution (<1.5% sigma). First, a two-loop write–verify protocol is proposed. Then, statistical experiments are conducted on RRAM array fabricated in Winbond’s 90-nm process. By incorporating the measured conductance distribution into DNN simulation considering the real weight mapping, inference accuracy with only 0.5% degradation over software baseline is achieved for CIFAR-10 data set even when 128 rows are read-out in parallel. By enabling parallel read-out, the system-level energy efficiency and throughput could be improved by $5.3 \times $ and $4.4 \times $ , respectively, compared to the 3-bit per cell RRAM used as MLC storage.
15 July 2021 in Neuromorphic Computing and Engineering
Focus in deep neural network hardware research for reducing latencies of memory fetches has steered in the direction of analog-based artificial neural networks (ANN). The promise of decreased latencies, increased computational parallelism, and higher storage densities with crossbar non-volatile memory (NVM) based in-memory-computing/processing-in-memory techniques is not without its caveats. This paper surveys this rich landscape and highlights the advantages and challenges of emerging NVMs as multi-level synaptic emulators in various neural network types and applications. Current and potential methods for reliably programming these devices in a crossbar matrix are discussed, as well as techniques for reliably integrating and propagating matrix products to emulate the well-known MAC-like operations throughout the neural network. This paper complements previous surveys, but most importantly uncovers further areas of ongoing research relating to the viability of analog-based ANN implementations based on state-of-the-art NVM technologies in the context of hardware accelerators. While many previous reviews of analog-based ANN focus on device characteristics, this review presents the perspective of crossbar arrays, peripheral circuitry and the required architectural and system considerations for an emerging memory crossbar neural network.
20 January 2021 in IEEE Transactions on Nuclear Science
This article compares and analyzes the single-event gate rupture (SEGR) response of silicon planar gate super-junction (SJ) power metal oxide semiconductor field effect transistors (MOSFETs) and vertical double diffused power MOSFETs (VDMOSs). When an incident heavy-ion strike is perpendicular to the gate oxide, the SEGR tolerances of SJ power MOSFETs (SJMOSs) and VDMOSs are similar. But, for heavy-ion strikes that are at different angles, SJMOS has better SEGR tolerance than VDMOS. This improved performance of SJMOS is due to the presence of an additional horizontal electric field component in SJMOS devices. This is validated using the experimental data and simulation results in this article.
29 October 2014 in IEEE Transactions on Nuclear Science
The impact of ionizing radiation on the retention and endurance of programmable metallization cells (PMC) ReRAM cells is investigated and presented for the first time, with additional work on resistance switching. This study shows that 60Co gamma-ray exposure has a minimal effect on the retention of PMC devices, up to a total ionizing dose (TID) of 2.8 Mrad (Ge30Se70), the maximum TID level tested. The retention of both high resistance states (HRS) and low resistance states (LRS) during exposure was tested. Endurance appears to be slightly reduced with gamma-ray exposure. The endurance was tested to maximum TID of 4.62 Mrad (Ge30Se70). DC response characterizations were also performed on PMC devices after cumulative dose exposures with 50 MeV protons and 100 keV electrons. The data show that PMCs are most sensitive to proton irradiation incident from the backside of the device. For the electron exposures, it is shown that the LRS is mostly unaffected, but the HRS drifts to lower resistance values with an increase in radiation exposure.
14 December 2018 in IEEE Transactions on Nuclear Science
Neuromorphic computing systems built with resistive random access memory (RRAM) are attractive solutions for implementing deep learning on-chip. In this paper, the single event and cumulative radiation damage susceptibility of a 1-transistor-1-resistor synaptic array architecture is investigated for the inference stage of a neuromorphic system. A physics-based SPICE RRAM model is developed to simulate the analog switching behavior of HfOx-based RRAM devices. SPICE simulations using the RRAM model are performed to capture the effects of transient heavy-ion strikes and experimental results obtained on GeSe-based RRAM devices are used to model the effects of cumulative radiation dose. Error patterns are fed into the software simulation of a multilayer perceptron, a representative artificial neural network for MNIST handwritten digit recognition. Radiation effects on the inference accuracy are analyzed. The results of the analysis indicate that the RRAM-based neuromorphic computing system is highly resistant to transient single-event effects due to the low cross section. The system is also shown to be tolerant to multi-Mrad levels of total ionizing and displacement damage dose.
9 November 2015 in IEEE Transactions on Nuclear Science
The Cu-SiO 2 based programmable metallization cell (PMC) is a promising alternative to the Ag-chalcogenide glass PMC because of its low power consumption and CMOS-compatibility. Understanding its total ionizing dose (TID) response helps in assessing the reliability of this technology in ionizing radiation environments and benefits its expansion in the space electronics market. In this paper, the impacts of TID on the switching characteristics of Cu-SiO 2 PMC are investigated for the first time. The devices were step irradiated with 60 Co gamma-rays to a maximum dose of 7.1 Mrad ( SiO 2 ). The results show that gamma-ray irradiation has a negligible impact on the virgin-state and on-state resistance of Cu-SiO 2 PMCs. The off-state resistance slightly decreases after the first 1.5 Mrad( SiO 2 ) of exposure, but this reduction saturates after higher levels of TID. Other switching characteristics such as the set voltage, multilevel switching capability and endurance were also studied, all of which did not show observable changes after gamma-ray radiation. The immunity to ionizing radiation is attributed to the suppression of the photo-doping process.