Talking to the assembled throngs at the tinyML Summit 2021, Raspberry Pi co-founder Eben Upton said the Raspberry Pi’s in house chip development team have been flat out improving the hardware’s machine learning skills.
Most of the work appears to be focused on developing lightweight accelerators for ultra-low power machine learning applications.
According to one of the slides he presented there will be three current generations of 'Pi Silicon' boards, two of which are from board partners, SparkFun's MicroMod RP2040 and Arduino's Nano RP2040 Connect.
The last one is from ArduCam which is working on an ArduCam Pico4ML that incorporates machine learning, camera, microphone and screen into the Pico package.
The last bullet point hints that the future chips might have lightweight accelerators possibly 4-8 multiply-accumulates (MACs) per clock cycle.