Akida is an implementation of a spiking neural network with pre-processing for both real-world signals and digital data. The spiking neural network is a type of artificial neural network that emulates the functionality of the human brain’s synapses and neurons. It can be trained using one-shot learning in supervised and unsupervised learning modes, making it applicable to a number of end-market applications including vision systems for civil surveillance, autonomous vehicles, robots and drones, financial technology and cybersecurity.
The Akida chip is relatively small in that is supports 1.2 million neurons and 10 billion synapses between neurons. But the small size makes it low cost and suitable for edge of network applications such as advanced driver assistance systems (ADAS), autonomous vehicles, drones, vision-guided robotics, surveillance and machine vision systems.
It includes sensor interfaces for traditional pixel-based imaging, dynamic vision sensors (DVS), Lidar, audio, and analog signals. It also has high-speed data interfaces such as PCI-Express, USB, and Ethernet. Embedded in the NSoC are data-to-spike converters designed to optimally convert popular data formats into spikes to train and be processed by the Akida Neuron Fabric.
The Akida chip can be networked together to allow scale up so that Akida can be used for complex neural network training and inferencing for many markets including agricultural technology (AgTech), cybersecurity and financial technology (FinTech).
The device is implemented in CMOS logic process but the company did not indicate what choices have been made about digital or temporal resolution for spike amplitude or timing? Nor did the company, which is fabless, indicate where the chips are being manufactured or in what manufacturing process.
Next: Big opportunity
The annual market for AI chipsets will surpass US$65 billion in 2025, according to market research firm Tractica (Boulder, Colo.).
“Despite their best efforts, no other company, large or small, has managed to bring a neuromorphic computing chip to market in production volumes,” said Lou DiNardo, BrainChip CEO, in a statement. “Akida, which is Greek for ‘spike,’ represents the first in a new breed of hardware solutions for AI. Artificial intelligence at the edge is going to be as significant and prolific as the microcontroller.” DiNardo added: “We are collaborating with major global manufacturers in a multi-market strategy to drive early adoption of the Akida NSoC.”
BrainChip is using the term NSoC to denote neuromorphic system-on-chip.
The company said that in comparison with leading convolutional neural network (CNN) accelerator ICs the NSoC is showing an “order of magnitude” performance gain in terms of images per second per watt when running industry standard benchmarks such as CIFAR-10 with comparable accuracy. The company did not say whether an order of magnitude meant a factor of two or of ten.
Spiking neural networks are inherently feed-forward dataflows, for both training and inference. Ingrained within the Akida neuron model are innovative training methodologies for supervised and unsupervised training. In the supervised mode, the initial layers of the network train themselves autonomously, while in the final fully-connected layers, labels can be applied, enabling these networks to function as classification networks. The Akida NSoC is designed to allow off-chip training in the Akida Development Environment, or on-chip training. An on-chip CPU is used to control the configuration of the Akida neuron fabric as well as off-chip communication of metadata.
The Akida Development Environment is available now for early access customers to begin the creation, training, and testing of spiking neural networks. The Akida NSoC is expected to begin sampling in 3Q19.
Related links and articles: