Opportunities for Analog Coding in Emerging Memory Systems
Authors:
Jesse H. Engel,
S. Burc Eryilmaz,
SangBum Kim,
Matthew BrightSky,
Chung Lam,
Hsiang-Lan Lung,
Bruno A. Olshausen,
H. -S. Philip Wong
Abstract:
The exponential growth in data generation and large-scale data analysis creates an unprecedented need for inexpensive, low-latency, and high-density information storage. This need has motivated significant research into multi-level memory systems that can store multiple bits of information per device. Although both the memory state of these devices and much of the data they store are intrinsically…
▽ More
The exponential growth in data generation and large-scale data analysis creates an unprecedented need for inexpensive, low-latency, and high-density information storage. This need has motivated significant research into multi-level memory systems that can store multiple bits of information per device. Although both the memory state of these devices and much of the data they store are intrinsically analog-valued, both are quantized for use with digital systems and discrete error correcting codes. Using phase change memory as a prototypical multi-level storage technology, we herein demonstrate that analog-valued devices can achieve higher capacities when paired with analog codes. Further, we find that storing analog signals directly through joint-coding can achieve low distortion with reduced coding complexity. By jointly optimizing for signal statistics, device statistics, and a distortion metric, finite-length analog encodings can perform comparable to digital systems with asymptotically infinite large encodings. These results show that end-to-end analog memory systems have not only the potential to reach higher storage capacities than discrete systems, but also to significantly lower coding complexity, leading to faster and more energy efficient storage.
△ Less
Submitted 21 January, 2017;
originally announced January 2017.
Training a Probabilistic Graphical Model with Resistive Switching Electronic Synapses
Authors:
S. Burc Eryilmaz,
Emre Neftci,
Siddharth Joshi,
SangBum Kim,
Matthew BrightSky,
Hsiang-Lan Lung,
Chung Lam,
Gert Cauwenberghs,
H. -S. Philip Wong
Abstract:
Current large scale implementations of deep learning and data mining require thousands of processors, massive amounts of off-chip memory, and consume gigajoules of energy. Emerging memory technologies such as nanoscale two-terminal resistive switching memory devices offer a compact, scalable and low power alternative that permits on-chip co-located processing and memory in fine-grain distributed p…
▽ More
Current large scale implementations of deep learning and data mining require thousands of processors, massive amounts of off-chip memory, and consume gigajoules of energy. Emerging memory technologies such as nanoscale two-terminal resistive switching memory devices offer a compact, scalable and low power alternative that permits on-chip co-located processing and memory in fine-grain distributed parallel architecture. Here we report first use of resistive switching memory devices for implementing and training a Restricted Boltzmann Machine (RBM), a generative probabilistic graphical model as a key component for unsupervised learning in deep networks. We experimentally demonstrate a 45-synapse RBM realized with 90 resistive switching phase change memory (PCM) elements trained with a bio-inspired variant of the Contrastive Divergence (CD) algorithm, implementing Hebbian and anti-Hebbian weight updates. The resistive PCM devices show a two-fold to ten-fold reduction in error rate in a missing pixel pattern completion task trained over 30 epochs, compared to untrained case. Measured programming energy consumption is 6.1 nJ per epoch with the resistive switching PCM devices, a factor of ~150 times lower than conventional processor-memory systems. We analyze and discuss the dependence of learning performance on cycle-to-cycle variations as well as number of gradual levels in the PCM analog memory devices.
△ Less
Submitted 9 October, 2016; v1 submitted 27 September, 2016;
originally announced September 2016.
