A fully autonomous car must act as a capable human driver and react in a timely and accurate manner in any situation to ensure maximal safety of the vehicle environment and driver. The vehicle must process data from a high number of vehicle sensors and run complex algorithms to build and respond to a dynamic perception model of the surrounding vehicle environment in less than 20ms. This requires enormous computing power. Furthermore, it is crucial that energy consumption is optimised to avoid computers that power self-driving being a huge driver of carbon emissions, and so ensure a green transition within the automotive industry.

AD requires a combination of AI and DSP to process all the data. To date, this has been handled by assembling multiple chips, resulting in high latency and power consumption. Additionally, current solutions can only use <20% of the maximum computing power due to architectural constraints. There are currently no commercially viable suppliers and manufacturers that can process the data required for L4/L5 AD.

It is broadly accepted that seven fundamental requirements are necessary to realise L4/5 AD: [1] Massive compute power efficiently delivered, [2] Very low latency, [3] Minimal energy consumption, [4] A combination of AI and DSP capabilities, [5] Deterministic processing, [6] In-field programmability, and [7] Affordable pricing.