BrainChip launches neuromorphic process for AI at the edge

BrainChip today announced the commercialization of its Akida neural networking processor. Aimed at a variety of edge and internet of things (IoT) applications, BrainChip claims to be the first commercial producer of neuromorphic AI chips, which could deliver benefits in ultra-low power and performance over conventional approaches.

As AI has continued to grow over the last few years, it’s expected that AI at the edge will become a bigger portion of the market. This is known as the artificial intelligence of things. Various conventional processor vendors such as Intel and Nvidia have launched AI chips for these lower-power environments, respectively, through their Movidius and Jetson product lines. Computing at the edge further results in lower latency than sending information to the cloud.

Neuromorphic computing and AI

BrainChip is taking a different approach with its neuromorphic processor. Neuromorphic computing differs from classical approaches to AI, which are based on convolutional neural networks (CNNs), by mimicking the brain much more closely through spiking neural networks (SNNs). Although neuromorphic chips are generally still digital, they tend to work based on asynchronous circuits, which means that there is no global clock. Depending on the specific application, neuromorphic can be orders of magnitude faster and requires less power.

BrainChip’s Akida architecture features such an event-based architecture. It can support on-chip training and inference, and it supports various sensor inputs such as vision, audio, olfactory, and smart transducer applications. BrainChip is targeting applications such as smart home, smart health, smart city, and smart transportation. Totaling 1.2 million neurons and 10 billion synapses, the chip scales from microwatts to milliwatts in power consumption. Akida is programmed through the MetaTF development environment.

With today’s announcement, BrainChip is starting to take orders for its first commercially available mini PCIe board featuring the Akida AKD1000 chip, with pricing starting at $500. It can be used as an embedded accelerator or co-processor. BrainChip notes that this completes its family of AKD1000 products, following the development kits that were introduced in October for x86- and Arm-based systems.

Akida on board with Mercedes

Akida is already featured in a notable product: the Mercedes EQXX concept car shown at CES earlier this month, where it was used for voice control to reduce power consumption by up to 10x.

BrainChip is not the only company working on neuromorphic computing, although it may be the furthest along in its commercialization. Of the bigger companies, Intel is most notable for its work on the Loihi 1 and 2 neuromorphic chips. Although Intel intends to commercialize its neuromorphic silicon at some point, currently it’s still in research through what Intel calls its Neuromorphic Research Community.