Analogue platform developed for artificial intelligence in edge devices

  

Based on TowerJazz’s commercial patented Y-Flash NVM on its 0.18um CMOS technology, the platform enables the development of ultra-low power Artificial Intelligence (AI) cores suitable for IoT edge devices and sensors such as fingerprint sensors, face and audio recognition applications among others.

According to the company, when compared with existing digitally based solutions this solution enables several orders of magnitude lower power consumption and is very cost effective due to the ability to implement it in less advanced technology nodes.

Teams from both TowerJazz and the Technion worked together on this project and the complete research paper will be published in the upcoming edition of Nature Electronics.

When compared to other NVM solutions targeting AI applications, the developed memristors are based on a commercial CMOS technology with enhanced reliability.

Single poly Y-Flash floating gate NVM transistors, originally designed for digital data storage, were converted into two-terminal analogue devices operated in the energy-efficient sub-threshold regime. The analogue memristors are tuned using optimised switching voltages and times to achieve 65 discrete resistive levels. These memristors are easily integrated with other devices on the same platform.

“We believe that analogue AI is the only solution for having AI in every edge device as this is the only way to meet the power consumption and cost requirements,” said Dr. Avi Strum, TowerJazz Sr. Vice President and General Manager of the Sensors Business Unit. “This is an important breakthrough development and a great solution that will enable the fast growth of the analogue AI market, aligned with market needs and existing and potential customer roadmaps”.