Stefan Mileschin

Last year we saw the announcement of Cadence’s Tensilica Q6 DSP IP which promised a new architecture that brings integration between vision DSP workloads and new optimised machine learning inferencing workloads. The addition of “AI” capabilities to existing DSP architectures bridges the gap between existing IP blocks such as CPUs or GPUs and more specialised dedicated inferencing IP blocks such as Cadence’s own Tensilica DNA100 block.

Today’s announcement is an evolution of last year’s Q6, further progressing the capabilities we saw introduced in the new architecture and enabling more performance, better density and better power efficiency.

Over the next few years Cadence sees significant growth opportunity for the vision DSP market, with the overall image sensor market growing at a rate of ~12% CAGR till 2025. Naturally those image sensors will need corresponding image processing power behind them in order to transform the raw image data into something meaningful. Particularly the automotive sector is projected to boom enormously in this regard with a continuous annual 36% growth rate, thanks to the projected need for dozens of sensors in future cars.

However the growth isn’t solely facilitated by the automotive sector. The mobile and smartphone sector is still projected to be the biggest market, and here growth opportunity is facilitated by the new trend of employing more and more camera modules in smartphones, something that over the last year in particular has become exceedingly evident. Other markets for opportunity are AR/VR headsets which also are projected to require a large numbers of cameras which will need image processing.

Here’s where the new Tensilica Q7 DSP comes into play. The IP is relatively straightforward in what it brings compared to its predecessor, and that could be summed up as a 2x increase in its performance capabilities.

https://www.anandtech.com/show/14337...-vision-q7-dsp