STMicroelectronics: From days to minutes: Fast bit-accurate MATLAB simulations with Catapult HLS MEX
Traditional bit-true modeling flows start from a MATLAB reference model and are then manually or semi-automatically translated into HLS/HDL. For Signal Array Processing applications (radar, sonar, medical imaging, etc.), this approach often becomes impractical: system-level simulations involve multiple DSP chains running in parallel, and non-linear operations prevent the use of vectorization techniques, making MATLAB simulations unbearably slow.
In this work, we show how generating MEX functions directly from Catapult HLS removes the simulation-time bottleneck while preserving hardware-accurate, bit-true behavior. This enables designers to run fast, hardware-faithful simulations directly in MATLAB and perform detailed accuracy and performance analyses without maintaining a separate high-level model.
A real-world case study on a Signal Array Processing ASIC will be presented, demonstrating the proposed methodology, the achieved speed-up, and its impact on the overall design flow.
Poznaj naszych ekspertów
Marco Castellano
Digital Design Manager
Marco Castellano earned his Laurea degree from the University of Pavia, Italy, in 2005. He continued his studies at the same university, completing a Ph.D. in electrical engineering in 2009. His doctoral research was a collaborative effort between the University of Pavia and STMicroelectronics, focusing on the creation of high-speed, low-power arithmetic circuits for Digital Signal Processing. In 2016, Marco advanced to a leadership role within an R&D team at STMicroelectronics, where he was instrumental in driving the innovation of embedded processing in analog products and sensors. His contributions to the field are documented in a series of published papers, presentations at scientific conferences, and a portfolio of patents, all centered around the integration of algorithms.
Luca Gandolfi
Digital HW Designer
Since 2018, Luca has worked at STMicroelectronics, specializing in the development of RTL‑consistent algorithmic models for DSP and TinyML. His work bridges high‑level modeling in MATLAB and Python with ultra‑low‑power ASIC implementations for smart sensors and Analog Front-Ends, including MEMS, infrared, and biomedical applications. Luca also contributes to the design and RTL implementation of dedicated digital signal‑processing IP cores.