Cadence DSP provides LLM and LVM AI at the edge

Want to search an image for the objects within it? Want to do this at the edge? Cadence has the solution!

In the first of two interviews with Cadence at Embedded World 2024, Guy chats with Suhas about AI/ML at the edge and Cadence’s highly configurable Tensilica AI DSP that eases and speeds the integration of AI into use cases for our daily lives.

Cadence’s Compute IP takes the hardware and software aspects of a design and creates an ecosystem – compilers, DSP, AI training environments, SDKs, etc. – that ensures the highest performance processing can be done with the power budget you are restricted by.

As Zahul explains, Cadence showed off several partner implementations of its technology for edge workloads like Large Language Models (LLMs) and Large Vision Models (LVMs). The latter combines visual- and text-prompt-based AI for some of the most powerful AI at the edge to date.

So what does an LVM look like in action? Well, we’re all familiar with the fact that an image can be analysed in real time by an AI. By using an LVM, however, you can also search the image for objects within it via a text prompt.

As we see from the demo – created in partnership with Labforge – any previously categorised object is detected and marked with a probability of correct identification. This model has been trained in the cloud with petabytes of data but is deployed for completely on-chip inferencing, hence edge AI.

If you are looking for next-level AI processing on chip and are constrained by power consumption, Cadence has you covered for everything you’ll need to get that implemented fast. Learn more about Cadence’s Tensilica Vision DSP by following the link to the board page below, where you can apply to evaluate this technology for your next commercial project.

Keep designing!