A Technology that Learns Like the Human Brain

Spiking Neuron Adaptive Processor (SNAP) is a technology that consists of many custom-designed cores that operate in parallel, making it significantly faster than software neural networks that run on a CPU or a Graphics Processing Unit (GPU). This new technology has the ability to learn autonomously, evolve and associate information just like the human brain. The technology is user configurable to meet a wide range of applications. 

SNAP is more energy efficient, enabling SNN to be integrated into portable devices for the local processing of sensor data. SNAP based neural networks can respond in real time with low latency, regardless of the neural network size. SNAP also implements learning rules in hardware, enabling Autonomous Features Extraction (AFE) directly from input data without the need for any software processing.

What Makes Our Technology So Exciting

What makes our technology exciting is that it learns from experience, autonomously like a human learns. It does not need to be trained with millions of samples like Deep Learning, it learns in seconds. Deep Learning networks need a power-hungry and huge supercomputer to train, and training takes days to weeks. BrainChip SNAP learns by itself, without specific training, and finds patterns in the input stream that humans may not be aware of. This rapid learning capability opens up a whole new area of possibilities, to find images in a long video, patterns in stock market data, and thousands of other applications where deep learning cannot be used.

The small footprint of SNAP, its low power requirements, and its high speed make it possible to embed the technology in a single chip that contains the entire solution, for instance in a camera chip. BrainChip SNAP is truly the next generation after Deep Learning.

How SNAP Works

The neurons we have developed autonomously learn through a process known as STDP (Synaptic Time Dependent Plasticity). Our fully digital neurons process input spikes directly in hardware and are all updated in parallel, which means that the response time of the network is independent of the network size. Sensory neurons convert physical stimuli into spikes. Learning occurs when the input is intense, or repeating through feedback at each neuron and this is directly correlated to the way the brain learns.

Depending on power consumption requirements, the SNAP technology can run at the speed of biological neurons or up to 1,000x faster. Software Neural Networks are limited in speed and complexity by the sequential processing method of a computer. SNAP operates completely in parallel with no dependence on software, which gives it its speed advantage.