Math/FFT Hardware accelerator like the NXP PowerQuad?

[etheory]

Hi there!

Are there any plans to add a hardware math accelerator in addition to the cordic and fmac that supports FFT's and other matrix operations entirely in hardware and runs in parallel to the CPU and communicates via DMA?

A good example of this is the NXP PowerQuad hardware: https://www.nxp.com/docs/en/application-note/AN12282.pdf.

I am using STM32 MCU's for audio synthesis/synthesizers and whilst the cordic/fmac is useful, having 2 cordics and 2 fmacs would be a LOT more useful, and even better, having a dedicated FFT hardware or vector processor with math capabilities that is independent of the main processor would be super useful.

Are there any plans to add something like this to your MCU lineup?

Thanks.

[Pavel A. · Accepted solution]

STM32s are not expensive. What if instead of waiting for ST to make a new chip you add a 2nd STM32 dedicated to your math workload? or 3rd?

For example, one my customer who needs redundant control, instead of a Cortex-R class MCU (pricey, hard to find) puts several normal STM32's with a cheap home-made arbitration circuit. They are happy.

[etheory]

Thanks @Pavel A. that's pretty much what I came up with as a solution, thanks. But it would still be better if the computation was possible in one chip due to latency. If it happens outside, you need to get the data between the IC's and back again. That requires memory, buffers, DMA etc. If it was on the one chip, none of that extra complexity is required. But sure, that's currently the only option. I've found a way to combine the CORDIC and FMAC to make a fourier transform processor, so I can have one cheaper chip do just that, whilst the main more expensive one does other tasks.

[Sarra.S]

Hello @etheory, 

As always, thank you for your insightful comments

All I can say is stay tuned