Breaking away from traditional sequential processing architectures, brain inspired computing mimics how the human brain is constructed, resulting in a massively parallel yet very low power consumption processor chip. The IBM / DARPA initiative in this area – referred to as SyNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable Electronics) – started in 2008. IBM’s TrueNorth NeuroSynaptic chip was developed as part of this program:
NeuroSynaptic processors take a radical new approach to military challenges including image recognition, machine vision, multi sensor fusion and signal processing. As an example, consider the conversion of a 480 x 360 pixel three colour image to a grey image, where the grey is calculated by adding the red, green & blue components together and dividing this by three. A traditional chip would take 480 x 360 steps to do this, which is 172,800 iterations. In comparison a each neutron of a synaptic chip would compute the grey value, taking only one iteration of parallel processing!
These chips are great at spotting patterns in data hence ideally suited to recognising sensor data in real time. For example, in a facial recognition app, one core of the chip might be focused on nose shape, one on hair texture and color, one on eye color, etc. Since each individual core is runs simultaneously the chip as a whole can perform this type of operation much more quickly and accurately.
Designed to process data “spikes” – rather than running to a fixed, fast clock cycle – the TrueNorth chip can operate for a week on an iPhone battery!
- Defence organisations are challenged to cope with an explosive increase in the amounts of data from a growing array of sensors often “at the edge” of the military network. The rate of growth of collected data outstrips the available bandwidth to move it around. This challenge is worse for the military as bandwidth is often severely constrained in certain deployments.
- So it makes sense for them to process the data as much as possible, as close to the sensor as possible and in real time. Classic, sequential computer architectures are not great at this. Stream Computing is better, but consumes much power and will be limited by the number of computers in the streaming array.
- In contrast, NeuroSynaptic chips are great at this highly parallel task and their low power consumption means that the application of advanced algorithms at the edge of network is possible.
There are no documented instances of the military using NeroSynaptic processing – it’s too early – BUT the non-military use cases have a direct military analogy, as illustrated in the following sections.
The “Jellyfish Robot” (shown below) monitors shipping lanes for safety, senses tsunamis and / or monitor sensitive marine environments. Similar “Roller robots” undertake search & rescue in dangerous environments using on board video cameras to looks for humans.
Military Analogy : Jellyfish Robot becomes an amazingly powerful, autonomous sonar bouy – detecting & classifying surface & subsurface targets. Roller robot could be equally as useful on the battlefield, either for search & recovery or counter insurgent activities.
“Vision Assist” emulates the visual cortex packaged into low-power, light-weight eye glasses designed to help the visually impaired. This could be multiple video and auditory sensors that capture and analyze this optical flow of data.
Military Analogy – the same application could be used to augment the sensory capability of the soldier or pilot of the future. Instead of making up for sensory failure, the same technology could be used for sensory augmentation increasing the effectiveness military personnel in undertaking their assigned task.
Sensors combined with neuromorphic chips the in future medical devices could recognize odors from certain bacteria. This is a hand-held thermometer for diagnosing minor illness or infection, It can smell what disease you have and notify you if a doctor visit is required.
Military Analogy – augmentation of wearable technology – which senses the soldier of the future’s wellbeing – with NeuroSynaptic capability will improve the military’s capability of getting the best out of each soldier, and better tailoring medical intervention after a trauma events.
The following short videos – produced by IBM Research – are the best way of finding out more about the SyNAPSE activities:
- New Architecture, TrueNorth & Compass & Live Demos
- Architecture, Neuron, Training & MNIST Example
- Corelet Conception, Development & Placement
- Scale down, Scale up & SyNAPSE University
I’m very interested in a discussion on this topic. Please leave me a comment, send me an e-mail or get in touch!