Catapult AI NN – From AI/ML Framework to Optimized RTL
As demand for Machine Learning increases, the need for custom hardware acceleration is exploding. Developing hardware that is optimally tuned for Performance, Power & Area (PPA) and optimized to the network and application delivers enormous competitive advantages in silicon. Designers working in AI/ML Frameworks such as TensorFlow, PyTorch have previously relied on “rule of thumb” or RTL design teams to assess the cost of implementation. Catapult AI NN now delivers a methodology and flow from “AI/ML Framework to RTL” enabling rapid exploration of network, quantization, design reuse factors on a layer-by-layer basis for PPA, generating production-ready custom RTL without needing extensive HLS knowledge or RTL skills.
Meet the speaker
Stuart Clubb
Technical Product Management Director
Stuart is responsible for Catapult HLS Synthesis and Verification Solutions since July 2017. Prior to this role, Stuart had been successfully managing the North American FAE team for Mentor/Siemens and Calypto Design Systems and was key to the growth achieved for the CSD products after the Calypto acquisition. Moving from the UK in 2001 to work at Mentor Graphics, Stuart held the position of Technical Marketing Engineer, initially on the Precision RTL synthesis product for 6 years and later on Catapult for 5 years. He has held various engineering and application engineering roles ASIC and FPGA RTL hardware design and verification. Stuart graduated from Brunel University, London, with a Bachelors of Science.
Related resources
StreamTV’s SeeCubic: Catapult HLS enables Ultra-D 3D without glasses
StreamTV's SeeCubic faced an impossible task: develop a real-time conversion IP block for a custom SoC without knowing the target technology. This IP was critical for their glasses-free 3D solution.
BLUEDOT: Accelerating NN-based DeepField-PQO design using Catapult HLS
Service providers face high encoding costs due to HQ videos demand. BLUEDOT offers AI-based DeepField-PQO filter improving coding efficiency. It uses HLS for fast filter implementation into IP targeted for FPGA, Catapult for ASIC.
Xperi®: A Designer’s Life with HLS
This webinar will discuss two aspects of their experience going from RTL to HLS. The first topic is using HLS for algorithms such as Face Detection th