There is a growing number of cases applying AI technology to semiconductors to achieve power efficiency at the edge. Microchip recently took steps to strengthen its AI-based edge solution capabilities by acquiring an AI technology company that achieves neural network optimization.

Microchip Technology announced on the 16th that it has acquired Neuronix AI Labs to strengthen its development capabilities for power-efficient AI-based edge solutions deployed on FPGAs.

Neuronics AI Labs is a company that provides neural network sparsity optimization technology to reduce the power, size, and computational load required for tasks such as image classification, object detection, and semantic segmentation while maintaining high levels of accuracy.

Microchip explained that this acquisition has enabled the more cost-effective development of large-scale edge deployments of components designed for computer vision applications in systems with cost, size, and power constraints, and has allowed for a multi-fold increase in the AI/ML processing capabilities of low-range and mid-range FPGAs.

Bruce Weyer, Vice President of Microchip’s FPGA business unit, said, “Through the technology of Neuronics AI Labs, we can improve the power efficiency of FPGAs and SoCs deployed in intelligent edge systems utilizing AI/ML algorithms. A design flow combining Neuronics’ technology with Microchip’s Vectorblox increases the performance efficiency of neural networks and also provides the outstanding GOPS/watt performance of Microchip’s low-power PolarFire FPGAs and SoCs.”

Vice President Bruce added, “System developers can now design and deploy architectures for small-footprint hardware that were previously difficult to implement due to size, heat, or power constraints.”
<bWith the acquisition of neuronics technology, even FPGA specialist developers who do not possess professional knowledge of FPGA design flows can utilize parallel processing capabilities using industry-standard AI frameworks.

By combining Neuronics' AI intellectual property with Microchip's existing compilers and software design kits, AI/ML algorithms can be implemented on customizable FPGA logic without the need for RTL (resistor-transition level) expertise or in-depth knowledge of basic FPGA architecture. Additionally, designs can be made to enable immediate CNN updates and upgrades without the need to reprogram the hardware.

"Neuronics AI Labs CEO Yaron Raz said, 'Neuronics AI Labs has focused on producing neural network acceleration architectures and algorithms that can completely change the size, power, performance, and cost-benefit of users.' He added, 'Joining Microchip gives us a unique opportunity to expand our FPGA portfolio, which has established the industry standard for power efficiency, and to integrate with these solutions.'"