fairfolio
Saved Projects

Brain-Inspired Circuitry for the Future of AI: Optimizing the Analog Response of RRAMs Under Pulsing for Synaptic Use

CSEF · 2019 Electronics & Electromagnetics First Award

Akhilesh V. Balasingam · Mitty (Archbishop) High School, San Jose, Santa Clara County

Overview

Objectives On tasks such as pattern recognition, the human brain, which consumes 20W of power and takes up 1.5L of space, substantially outperforms artificial intelligence (AI) algorithms running on large clusters of von Neumann (vN) computers. To close this glaring gap, intense research is underway on brain-inspired alternatives to vN. Like the brain, these alternatives capture the computationally-intensive parts of AI algorithms in massively parallel circuitry, interweaving low-power computing and analog memory (synapse) elements. My research optimizes the analog response of resistive random access memories (RRAMs), to permit their use as synaptic elements in brain-inspired architectures for AI-centric computing. I chose to study non- filamentary (nf) RRAMs, because their conductance can be adjusted in a continuous, or analog, fashion with voltage pulses applied between their 2 terminals. Methods Key steps of my research: (a) Write a Kinetic Monte Carlo simulator (KMC), capturing mixed ionic- electronic conduction central to nf-RRAM operation. Model conductance using nonlinear resistor network-- apply Kirchhoff's Laws and Newton's Method. (b) Use KMC to study nf-RRAM response under different pulsing schemes. (c) Find pulsing schemes that yield the best synaptic response. (d) Benchmark the system- level accuracy of my synaptic designs in NeuroSim, a neural network simulator. Results My contributions: (a) Developed physics-based KMC simulator in Python. (b) Validated KMC by comparing its predictions with published experimental data. (c) Using KMC characterized the synaptic behavior of nf-RRAMs under several pulsing schemes: standard, stepping, hybrid. (d) Standard pulsing yielded the poorest synaptic behavior. (e) Hybrid scheme yielded the most linear/symmetrical synaptic response. (f) Analyzed the system-level behavior of my synapses in NeuroSim and showed that my hybrid scheme yields nearly-ideal learning accuracy. Conclusions I studied and optimized the synaptic behavior of nf-RRAMs, using a KMC simulator I developed. Using my simulator, I identified a pulsing scheme that yields synaptic behavior with the desired analog properties of linearity and symmetry, and at the system-level I showed that this scheme yields 90% accuracy, which is very close to the 93% rate achieved by the ideal synapse.

Summary statement

I developed a hybrid pulsing scheme that allows nf-RRAMs to be used as synapses in massively parallel brain-inspired architectures, which can enable AI to be performed on small IoT (Internet of Things) devices.

Help received

I developed and performed the simulations and analysis on my own. I would like to thank my mentor for guiding me through the current literature. I would like to thank my school math teacher for helpful discussions and encouragement.

Awards (1)

Competition history

  • CSEF 2019 Electronics & Electromagnetics · Entry S1004

Resources

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

Source: California Science & Engineering Fair public projects

Save projects to your library

Sign in with Google to keep track of projects you find interesting, organized into folders. An account also raises your daily allowance for “Has this been done?”, and lets you create a key for the MCP server with a much higher limit than anonymous use. Browsing stays public.

Continue with Google