Enabling a solution for on-device and edge AI processing is about more than providing raw TOPS in an SoC. In the fast-evolving world of AI, solutions must provide both high performance and high utilization while handling many more “irregular” operations and not just matrix multiplies (transformers, LSTM, etc.), do so within a low-power and small-area profile with minimal accesses to memory, and be easy to use by developers for the networks of today and of the future.
In this presentation, we will discuss Cadence’s AI IP products enabling ultra-low-energy, battery-powered devices up to high-end applications requiring many hundreds of TOPs, supported by powerful software tools that enable a no-code environment for mapping networks to target executables.
