Analog ASIC Artificial Intelligence at the Edge

Enabling Analog Artificial Intelligence by the systematic generation of Analog Neural Networks from well-known Artificial Intelligence software tools. The circuits and methodologies developed here enable “AI at the Edge” meaning local, low power, analog AI that provides, for example, medical devices that analyse signals to asses the need for intervention; voice recognition devices that do not send recordings to the internet; smoke detectors that recognise the chemical composition of gas in the air and similar. All this is achieved in low cost manufacturing using standard CMOS fabrication steps. The innovation is the ability to configure many thousands of small analog computation units into a neural network. Access to this technology is simplified: the customer need not understand how the analog core is operating. Industry standard AI tools generate configuration files that are read to define the AI processing. The system includes data acquisition elements that hook directly to the analog AI core. This enables, for example, a complete EEG chip including 64 low noise amplifiers and ADC convertors to feed into a trained analog neural net that recognises onset of seizure in epileptic patients or pain in neo-natal children.

Faculty Supervisor:

Mohammad Hossein Zarifi

Student:

Partner:

SiliconIntervention Inc.

Discipline:

Engineering

Sector:

Professional, scientific and technical services

University:

The University of British Columbia - Okanagan

Program: