BrainChip machine learning chip was successfully tape-out on 22nm FDSOI

BrainChip has successfully taped out the AKD1500 machine learning chip on GlobalFoundries' 22nm FD-SOI process.

This reference design is a key milestone and is part of validating BrainChip IP across different processes and foundries. BrainChip has been working with Renesas to integrate an IP called Akida into the chip that was taped out last month.

The AKD1500 uses the latest features of BrainChip AI IP, combined with the ultra-low leakage FDX22 process, and is suitable for always-on sensor applications or other low-power embedded applications in AIoT, industrial, consumer and automotive markets.

For AI where the sensor is always online, Akida can offload the feed-forward network without using the CPU, resulting in extremely energy-efficient smart sensors. For other low-power embedded applications, where the CPU alone performs the sensor fusion process or runs the application, Akida manages all AI acceleration to significantly reduce CPU workload and system load, freeing the MCU for other general-purpose computing.

AKD1500 accelerator reference chip, using quad/octal SPI interface for MCU modules and PCIe for expandable cards, which helps partners develop and demonstrate their solutions as a way to integrate Akida IP into their production SoC in the foundation.

"This is an important milestone in the continued innovation of our event-based neuromorphic Akida IP, demonstrating that it is fully digital and portable across foundries," said Anil Mankar, co-founder and chief development officer of BrainChip. "The AKD1500 reference chip uses GlobalFoundries The low-power FD SOI platform demonstrates the possibilities of smart sensors in edge AI.”