Energy efficiency is all anyone talks about these days: rightfully so. As we squeeze as much juice as we can extract from this planet, to thrive on our survival, we must then find sustainable solutions that can satiate a global regime based on growth.
Now, to begin this article I should explain the titled quote as it has inspired Hoursec’s approach to enabling small intelligent devices capable of achieving real-time inference computation, with charging / updating cycles of up to a year to give customers some space and clouds, some commitment. Without going into the philosophy of moving systems, saying “computation does not require energy, only the movement of data” is along the same line of logic as saying “smashing an apple on the floor does not require throwing it down, only the application of pressure from the floor to the apple”; in practically it says that there are multiple ways to innovate on the computation required for machine intelligence. Alexandra says that Hoursec’s approach innovates on mathematical computation through their algorithms related to control theory instead of optimisations. What this means is that they are improving the management of traffic driving through a city by innovating on the traffic-light system instead of the size or speed limit of the roads; only for hoursec the city is the chip, the traffic is data, and, their product is the traffic-light management system. Hoursec’s computational improvements come in the form of how much energy you need to put into a system using their AI-on-chip platform, and, how the data in that system is being managed by their platform.
Alexandra’s approach is backed by a string of research that reached her through working with professors from the University of Berkely, California, specifically research dating back to Leon O. Chua, inventor of the memristor in the early 1970’s – Professor Leon O. Chua established the mathematical foundation for non-linear circuit theory which is a key theme amongst the trend of brain-inspired computing.
The team at hoursec are building their core products to enable of a future with Artificial Intelligence of Things; a word play on Internet of Things that envisages a future where small devices (or all devices) are intelligent (that is can capture information around them, as an input, and transmit informed inferences to other devices based on the task that they are built to achieve). Echoing the introductory paragraph, energy efficiency is the biggest barrier to realising this future – Hoursec’s software innovation (their traffic-light management system) achieves energy reduction of 1000x whilst their hardware innovation reduces it by a further factor of 3x. Alexandra’s aim is to serve the markets for sensors, focussing first on vision recognition (one of the biggest applications for large algorithmic models in AI alongside Natural Language Processing).
For the moment serving the market for autonomous driving is in the middle-distance – with car companies such as Tesla designing their own tailor made chips, the autonomous driving race is still up for grabs. But for the time being, Hoursec can be considered a leading light in Europe’s efforts to enable intelligence at the edge of the network.