MathWorks launches new support package for Qualcomm Hexagon NPU

MathWorks has announced a new hardware support package designed for the Qualcomm Hexagon Neural Processing Unit (NPU), integrated into the Snapdragon series of processors.

The package facilitates automated code generation from MATLAB and Simulink models, specifically optimized for the Hexagon NPU architecture. This development aims to enhance data accuracy and standards compliance while improving productivity in developer workflows.

"Qualcomm Technologies' mission is to empower engineers to create innovative products and applications that leverage the full potential of Snapdragon processors," said Vinesh Sukumar, Senior Director of Product Management at Qualcomm Technologies.

"As the demand for high-performance NPU applications continues to grow, the collaboration between MathWorks and Qualcomm Technologies enables developers to efficiently develop embedded applications optimised for the Hexagon NPU," Sukumar said.

The new hardware support package by MathWorks addresses significant challenges faced during the development process. These include reducing compilation errors, eliminating the reliance on hand-written frameworks, simplifying manual numeric verification, and overcoming the challenge of tracing code back to the model.

The package allows engineers with little or no expertise in the Hexagon NPU toolchain to perform Processor-In-the-Loop (PIL) simulations directly in Simulink by leveraging the Hexagon simulator. This ensures that the optimized target code functions correctly and efficiently before deployment on actual hardware.

"By harnessing the power of Model-Based Design and the advanced capabilities of the Hexagon NPU, we are making it easier for engineers to create high-quality NPU applications," said Jim Tung, MathWorks Fellow. "Our collaboration with Qualcomm Technologies is focused on removing manual processes in favour of highly automated workflows to help engineering organisations reduce time to market and ensure optimal performance."

MathWorks' MATLAB and Simulink are widely recognized for their automatic code generation capabilities. These tools are already optimized for a variety of processors, including ARM Cortex-M, ARM Cortex-A, ARM Cortex-R, Infineon AURIX TC3x and TC4x, as well as GPUs from ARM and NVIDIA. MATLAB and Simulink are commonly used in industries such as automotive, aerospace, and medical devices for algorithm development, data analysis, visualisation, and numeric computation.

The introduction of the new hardware support package signifies a joint effort by MathWorks and Qualcomm Technologies to streamline the development of high-performance neural processing applications. This new tool aims to expedite the development cycle, reducing the time required to bring new innovations to market, while maintaining high standards of accuracy and compliance.