New on GroupGets: GAP8

Greenwaves Technologies is making their grand entrance onto GroupGets with the GAPUINO Combo Kit, sporting the new groundbreaking GAP8 processor. Including all of the essential hardware and an SDK, this kit has everything you need to begin developing with the GAP8. Since this technology is so new, we reached out to Greenwaves for a Q&A to see what the GAP8 is, and what it can do.

GAP8 is a product focused on bringing machine learning and signal processing on rich data sources (images, sounds, vibrations, radar signals, etc) to battery operated devices. In many IoT applications such as wearables or edge sensors power is extremely limited yet battery life, which needs to be measured in weeks or years, is very important. Compared to current solutions GAP8 delivers a 20 times improvement in energy usage for inference operations combined with an MCU-like functionality for standby & communications.

GAP8 is a programmable device which can be used to reduce the energy consumption of any algorithm that can be parallelized. This includes algorithms such as Fast Fourier Transforms (FFTs), Filters, Neural Network layers and many more.

GAP8’s open source SDK includes many examples which allow customers to quickly get up to speed developing applications on GAP8.

In many edge applications inference is a significant operation but often a lot of signal processing needs to be done before it can be carried out. GAP8 can reduce the power consumption of not just AI algorithms but also many signal processing ones.

In many cases applications were possible, but not on a battery. Adding battery operation allows sensor installation costs to be reduced making deployment possible in a much wider range of cases. For example, counting the number of people using a meeting room is a very interesting application for facilities management. Knowing how a meeting room is being used gives a facilities manager the ability to plan how many meeting rooms to provide and how large they should be. However, this business case is destroyed if sensors require expensive cabling.

Like most devices in the IoT space we do not see one ‘killer app’. In the semiconductor world this would be a single app generating sales of hundreds of millions of units. We see a wide range of use cases spanning smart building, agriculture & cities, wearables and industry 2.0.

At present we are working on a unified API across all supported operating systems for GAP8. We are calling this API PMSIS. We are also working on an evolution of the AutoTiler tool that will further optimize memory movement in GAP8 and adds support for optimizing allocation of memory for activations and features in neural networks.

We are working on a series of ‘proof of concept’ boards for GAP8 integrated with several different sensor types: higher resolution, global shutter cameras, infra-red cameras, radar, etc.