Siemens Digital Industries Software has unveiled a new software for high level synthesis (HLS) of neural network accelerators for applications specific circuits (ASICs) and system-on-chips (SoCs).
Siemens said the software, called Catapult AI NN, is a solution that starts with a neural network description from an AI framework then converts it into C++ and synthesizes it into an RTL accelerator in Verilog or VHDL for implementation in silicon.
The software was developed with U.S. Department of Energy lab Fermilab and other contributors to address machine learning accelerator design for power, performance and area on custom silicon.
As AI and machine learning tasks migrate from data centers to consumer appliances and healthcare and other devices, AI hardware needs to minimize power consumption, lower cost and maximize end-product differentiation, Siemens said. However, engineers are more comfortable working with tools like TensorFlow, PyTorch or Keras rather than synthesizable C++, Verilog or VHDL. Catapult AI NN offers a bridge between the technologies to accelerate machine learning applications for ASIC and SoC.
Additionally, Catapult AI NN extends the capabilities of hls4ml to ASIC and SoC design with a dedicated library of specialized C++ machine learning functions that are tailored to ASIC design.
"Particle detector applications have extremely stringent edge AI constraints," said Panagiotis Spentzouris, associate lab director for emerging technologies at Fermilab. "Through our collaboration with Siemens, we were able to develop Catapult AI NN, a synthesis framework that leverages the expertise of our scientists and AI experts without requiring them to become ASIC designers. Moreover, this powerful new framework is also ideal for seasoned hardware experts."
