Digital Signal Processors (DSPs)

Whilst DSPs are mostly found in mobile platforms, they are an important form of accelerator.

A DSP is typically made up of a number of parallel special purpose blocks and are most useful in applications where a large number of mathematical operations need to be performed quickly and repeatedly on a data series. Whilst they were originally designed to handle typical analog data flows in activities such as wireless transmission and audio they are very efficient at operations that are the building blocks of most AI processing pipelines such as convolution.

There are specialist forms of DSP for individual markets:

  • Video processing

  • Audio processing

  • Neural networks

However, DSP elements are most frequently found inside SoCs or as hard IP with an FPGA. See also: