This AI processor consumes less than 1 milliwatt of power. Presented by BrainChip Akida Pico

BrainChip has unveiled the ultra-low-powered neural processor (NPU) Akida Pico, which consumes less than 1 milliwatt!

This NPU can be used as an AI accelerator for very compact, battery-powered smart devices. For example, hearing aids, headphones with noise cancellation and various medical equipment. The Akida Pico NPU can be implemented as a standalone chip or incorporated into other chips.

Akida Pico is an NPU that contains only 150,000 transistors and does not contain its own memory. The chip combines the NPU, DMA, and AXI bus matrix to connect to other parts of the processor or microcontroller.

Interestingly, the chip is manufactured at the capacity of GlobalFoundries on the 22FDX process technology (22 nm, FD SOI), and at the same time it consumes incredibly little.

More specifically, the Akida Pico NPU consumes less than 1 milliwatt with a crystal area of only 0.12 mm2. If you add 50 KB of SRAM memory to the chip, which is possible, the crystal size increases to 0.18 mm2. Importantly, the company says the core can be synthesized for any manufacturing facility, including low-cost 12nm TSMC nodes and ultra-high-performance 3nm class processes if needed. In the latter case, consumption is likely to be several times lower.

The company does not directly talk about productivity, but it is clear that the indicators should be extremely modest.