To address these challenges more effectively, Microchip Technology has introduced the VectorBlox™ 3.0 Accelerator Software Development Kit (SDK), which simplifies the implementation of FPGA-based AI and accelerates its time to market. The VectorBlox 3.0 SDK, provided free of charge to developers along with the corresponding CoreVectorBlox IP, is designed as an integrated toolchain that streamlines the optimization, compilation, and deployment of convolutional neural network (CNN) models on FPGA-based platforms and PolarFire® SoCs. Because the accelerator efficiently scales to models of varying sizes and supports multiple AI workloads on a single device, customers can integrate various vision- or sensor-based AI functions onto a single, low-power FPGA.

“As the complexity of AI models continues to increase, compression is becoming essential for implementing edge intelligence,” said Shakeel Peera, corporate vice president and general manager of Microchip’s FPGA business unit. “VectorBlox 3.0 leverages scatter-based model compression from Neuronix, a company acquired by Microchip, to reduce computing requirements while maintaining accuracy.”.

Thanks to its compatibility with sparse neural networks, VectorBlox 3.0 facilitates the efficient execution of vision-based CNN models by omitting zero-value operations. This capability helps developers accelerate inference performance and reduce power consumption, a significant advantage for always-on AI applications at the edge, which must balance response speed and energy efficiency. The use of sparse-based model compression is designed to decrease computation and memory requirements while maintaining accuracy.

“VectorBlox acceleration on Microchip’s PolarFire SoCs enables the efficient deployment of advanced onboard AI channels for low-latency payload operations in orbit,” said Vito Fortunato, SPACEDGE™ Services Line Director at Planetek Italia.

“This platform has allowed us to validate real-time Earth observation processing functionalities such as object detection, semantic scene analysis, and actionable information products generated at the edge that have been installed on the AI-eXpress-1 satellite, launched in 2025, in addition to providing the radiation resilience and operational reliability needed for continuous operations in low Earth orbit.”

In addition, Spacecraft Pose Network v2 (SPNv2), a neural network designed to estimate position and orientation using vision data, enables autonomous navigation and proximity operations in space for applications such as autonomous approach and docking, space debris removal, satellite inspection, and training flights. The solution, which utilizes mid-range PolarFire FPGAs and SoCs with efficient power consumption and single event immunity (SEU), provides secure boot, tamper protection, and high reliability in harsh environments. These features are essential for mission-critical defense, aerospace, and industrial applications where long operational life, data protection, and system resilience are paramount.

“The combination of the PolarFire and VectorBlox SoCs creates a powerful synergy for the full-orbit deployment of AI-based autonomy solutions,” said Federico Fontana, Head of Hardware Engineering at AIKO. “We have seen this throughout the deployment of our clear CHARLES, which provides cloud connectivity and spacecraft detection for adaptive and autonomous payload operations on energy-efficient platforms, representing a further step towards increasingly autonomous, reactive, and software-defined space systems.”

VectorBlox SDK v3.0 features support for Microchip's Libero® SoC Design Suite and is compatible with CoreVectorBlox IP