Weights Safety Mechanism In An Artificial Neural Network Processor

An apparatus for detecting errors in weights used in a neural network processor, comprising:

• a memory configured to store blocks of n weights, each block of weights protected by and having a corresponding first cyclic redundancy code (CRC) checksum stored in said memory with said weights, each first CRC checksum calculated a priori over a block of n weights at run-time, said weights used in neural network calculations on said processor;
• one or more CRC engines, each CRC engine operative to:
• receive a block of weights retrieved from said memory and to calculate a second CRC checksum over said weights;
• and
• verify whether said second CRC checksum matches said first CRC checksum retrieved from said memory for a particular block of weights and generating an error if a mismatch is detected.

A method of detecting errors in weights used in a neural network processor, the method comprising:

• parsing a model of a neural network and determining a CRC block size n for each layer of the neural network;
• calculating a CRC checksum value for each block of n weights, said CRC checksum values operative to protect the weights in each block;
• populating a memory with blocks of weights including associated precalculated CRC checksums;
• configuring a layer control unit (LCU) circuit for each layer of the neural network to skip over precalculated CRC checksums in each block during neural network calculations;
• wherein said weights are subsequently read from the memory and used in neural network calculations performed on said processor;
• and
• wherein said CRC checksum values are subsequently read from the memory and compared to CRC checksums calculated over weights read from the memory for each block and used in the neural network calculations.