VectorBlox 1, 2 and 3 are evolutionary generations of Microchip Technology’s VectorBlox Accelerator Software Development Kit (SDK) and associated CoreVectorBlox IP. They are software-to-hardware ecosystems for compiling, running and optimising machine learning/AI models on low-power PolarFire FPGAs and SoCs.
The latest release, VectorBlox 3.0 SDK, is offered free of charge to developers, designed as an integrated toolchain alongside the CoreVectorBlox IP, to streamline the deployment of convolutional neural network (CNN) models. Because the accelerator scales efficiently across model sizes and supports multiple AI workloads on a single device, developers can consolidate various vision or sensor‑based AI functions on a single low-power FPGA.
Microchip has used expertise from Neuronix AI Labs, the company it acquired in 2024. Neuronix brought advanced sparsity-based model compression technology, for reduced compute, power and thermal demands of AI algorithms such as CNNs.
“As AI models continue to grow in complexity, compression is becoming essential for deploying intelligence at the edge. With VectorBlox 3.0, we’re leveraging sparsity-based model compression from our Neuronix acquisition to reduce compute demands while preserving accuracy,” said Shakeel Peera, corporate vice president and GM of Microchip’s FPGA business unit.
With support for sparse neural networks, VectorBlox 3.0 helps enable efficient execution of vision-based CNN models by skipping zero‑valued operations. This capability helps developers accelerate inference performance while reducing power consumption, an important advantage for always‑on edge AI applications that must balance responsiveness with energy efficiency. Enabling sparsity-based model compression is designed to reduce compute and memory demands, while preserving accuracy.
