FPGAs increase memory and DSP for always-on IoT design

The configurable, energy-efficient FPGAs can be used in mobile heterogeneous computing (MHC), such as smartphones, wearables, drones, 360 cameras, human-machine interfaces (HMIs) and industrial automation, as well as security and surveillance products.
This member of the iCE40 Ultra family delivers eight times more memory (1.1Mbit RAM), twice the digital signal processors (with eight DSPs), and an increased number of I/Os compared with the earlier iCE40 Ultra and iCE40 UltraLite.
I/O includes SPI, UART, I²C, I³C to deliver multiple signals over a single PCB trace, which eliminates routing issues, reduces system cost and simplifies designs, says the company. The scope of I/O makes the FPGA suitable for voice recognition, gesture recognition, image recognition, haptics, graphics acceleration, signal aggregation or I³C, the MIPI standard with higher bandwidth, but which is backwards-compatible with I²C. The flexible I/Os also enable a more distributed heterogeneous processing architecture, allowing for design flexibility in introducing innovations such as always-on sensor buffers and acoustic beam forming.
MHC is concentrated around an energy-efficient method for computing algorithms quickly and locally, offloading functions from the processor, while keeping the ed product in always-on state to instantly process command locally, rather than sending to the cloud. Smartphones and IoT edge products, such as wearables and home audio assisted devices can benefit from this always-on, instant response capability.
The increased number of DSPs means that higher quality algorithms can be processe, and the increased memory capacity allows data to be buffered for longer low-power states.
In addition to always-on sensor buffer and distributed processing for mobile devices at sub-1mW power consumption, and always-on sensor functionality while the application processor is in sleep mode, some of the innovations envisioned are gesture detection, facial recognition, audio enhancement, audio beam forming, phrase detection, double tap, shake-to-wake and pedestrian dead reckoning (PDR). Frame buffer and graphics acceleration is possible, for wearables and white goods, with an always-on display while the application processor is in sleep mode. Multi-layer graphics acceleration improves system power consumption and multiple microphones improve audio quality to reduce background noise as well as for audio equalisation.
Package sizes start at 2.15 x 2.55mm and a QFN package for industrial sectors. Evaluation samples and boards are available now.