BrainChip Holdings, which describes itself as “a leading provider of ultra-low power, high-performance AI technologies and the world’s first commercial provider of neuromorphic AI chips,” today announced that it has begun accepting orders for the first commercially available PCIe add-in board powered by an Akida Neural Network Processor.
The AKD1000-based board can be connected to an existing developer system to open up opportunities for a wide range of AI edge applications, including Smart City, Smart Health, Smart Home and Smart Transportation. BrainChip will also offer system integrators and developers complete layout files and Bill of Materials (BOM) so they can build their own boards and implement AKD1000 chips in high volume as a standalone embedded accelerator or coprocessor.
The 40 x 76mm board houses a processor, 512MB LPDDR4-2400 memory, and 128Mbit Quad SPI NOR flash. In addition, there is a built-in Akida core current monitor and two GPIO lines. The card is equipped with a PCIe 2.0 x1 interface. Its advantages include high performance, security, low power consumption and the ability to conduct AI training on the device itself, regardless of the cloud. The price starts at $499.
In October last year, the company introduced two variants of developer kits. One is based on an x86-compatible Shuttle mini PC, and the other is based on an ARM-compatible Raspberry Pi single-board microcomputer.