Jotrin Electronics
Description Quantity Total (USD) Operation
Shopping cart products
Shopping cart products : 0
Home > Energy collection > Mentor launches AI HLS development kit and updates Calibre's AI capabi

Mentor launches AI HLS development kit and updates Calibre's AI capabilities

Published time: 2019-12-20 11:08:54

Mentor, A Siemens Company has announced an artificial intelligence/machine learning development kit and added AI/ML enhancements to both tools to help customers bring smarter AI/ML chips to market faster.

The company's Catapult Software Advanced Synthesis (HLS) toolkit and ecosystem are designed to help customers quickly launch the development of complex machine learning IC architectures. At the same time, Mentor added AI / ML infrastructure to the entire Calibre platform and introduced two AI / ML technologies, namely Calibre Machine Learning OPC (mlOPC) and LFD with Machine Learning. Both technologies make machine learning faster and more accurate.


The Catapult HLS AI toolkit helps developers use AI/ML-based accelerators for edge applications. Based on HLS C++, the kit provides object detection reference design and IP to help designers quickly find the optimal power, performance and area implementation of a neural network accelerator engine. The solution also includes a complete setup for building an AI / ML demonstrator platform that provides real-time HDMI feeds on the FPGA prototype board.

At the same time, the new Calibre mlOPC product has a threefold increase in OPC runtime compared to existing tools. It does this through intelligent feature extraction and machine learning algorithms to predict OPC output to nanometer accuracy, eliminating up to 75% of OPC runtime.

Mentor also added machine learning options to its lithography simulation tools. This new feature is designed to achieve high accuracy and high block performance as well as full chip analysis. The predictive function of this function focuses on the high-risk layout mode of detailed lithography simulation, eliminating risk from this computationally intensive step. The results show a 10 to 20-fold improvement in performance compared to simulations based on full-chip models while maintaining optimal accuracy.



Account Center


Live Chat