2021 BRN Discussion, page-32314

https://sbir.nasa.gov/SBIR/abstracts/20/sbir/phase1/SBIR-20-1-H6.22-6021.html
PROPOSAL NUMBER: 20-1- H6.22-6021
SUBTOPIC TITLE: Deep Neural Net and Neuromorphic Processors for In-Space Autonomy and CognitionPROPOSAL
TITLE: Radiation-Tolerant Neuromorphic Processor Featuring Nonvolatile Compute-in-Memory Architecture
Technical Abstract (Limit 2000 characters, approximately 200 words)

Low power and high speed neuromorphic processors have an on-demand need for the growing edge-AI market. Non-Volatile Memory (NVM) based compute-in-memory architecture using Flash memory, STT-MRAM or ReRAM has shown promising results for high energy efficiency compared to the traditional computing architecture. While the various technical challenges such as slow access speed and high fabrication cost exist, radiation-tolerance is the key merit of using emerging nonvolatile memories such as STT-MRAM or ReRAM against Flash memory for space applications.

In this SBIR phase I project, we plan to investigate key reliability architectural challenges, and solutions such that these radiation-tolerant NVMs can be deployed in compute-in-memory based neuromorphic processors for higher performance and energy efficiency compared to the conventional general purpose processors. With this objective, we propose to explore suitable micro-architecture of NVM based compute-in-memory processor, design an NVM based neuromorphic core, and optimize a neural network architecture addressing variation and reliability challenges of the NVM cells along with model quantization.

Potential NASA Applications (Limit 1500 characters, approximately 150 words)High speed vision processing from satellites requires massive computing power for running deep neural networks. Emerging nonvolatile memory based neuromorphic processing cores will be able to pave the way for such high speed computing capabilities with suitable radiation-tolerance that the space environment requires.


https://www.f6s.com/anaflash-inc
ANAFLASH Inc. is an NSF, DoD, NASA and Berkeley SkyDeck funded Silicon 100 startup that develops a zero standby power non-volatile processor using a standard logic process for smart edge AI devices.

Our low power embedded non-volatile memory does not need any process overhead from a silicon foundry, which enables a cost-effective and zero standby power non-volatile processor that can run wireless IoT devices power efficiently extending their batter power significantly.

Along with our proprietary computing-in-memory based processor architecture solution, our edge AI processor can provide the order-of-magnitude improvement of the performance and the power efficiency compared to traditional Von Neumann architecture based processors.


WO2021158512A1NEURAL NETWORK UNIT

This neural network unit includes: a main synapse array having a plurality of main synapses in rows and columns of the array, each connected to one or more bit lines associated with the network unit; a main synapse driver for applying an input to one or more of the main synapses in a row; a reference synapse for generating a particular output to a selected one or more associated bit lines; and, a sensing circuit for controlling the reference synapse in generating the output in a selected column of the array, wherein the sensing circuit is adapted to infer an output from one or more main synapses in a selected column by: defining a reference range of the output to be generated by reference synapse; setting a reference level from the reference range; activating a selected reference synapse to implement the reference level that is set or adjusted; and comparing an output from the main synapse summed with the output from the activated reference synapse in the selected column with a threshold value predetermined.


Anaflash uses ReRAM/Memristor tech - analog.