Ultra-Compact FPGA for AI

Proc10N: Compact FPGA for AI and High-Performance Embedded Systems

The Proc10N™ is a compact FPGA for AI module powered by the Altera Stratix® 10 NX FPGA. It delivers exceptional performance for AI acceleration, compute-intensive processing, and low-latency edge systems. With integrated Tensor Blocks, HBM2 memory, and high-bandwidth data interfaces, the Proc10N offers unparalleled throughput for high-speed embedded designs. Moreover, its compact form factor ensures seamless integration into advanced embedded platforms while maintaining maximum compute capability.

Performance and Memory in the Proc10N Module

Each Proc10N module integrates HBM2 memory, providing 10× higher DRAM/SRAM bandwidth compared to discrete DDR4 and QDR technologies. It also delivers 143 INT8 TOPS / FP16 TFLOPS through dedicated Tensor Blocks and offers 1,600 Gb/s of customizable I/O for extreme data throughput. Consequently, the Proc10N delivers industry-leading compute density at a competitive SWaP profile. Additionally, its robust hardware architecture ensures reliable operation in demanding, mission-critical environments and provides a long-term, cost-efficient solution for complex embedded systems. Therefore, it is ideal for applications requiring deep learning, real-time inference, high-bandwidth acquisition, and deterministic system behavior.

Real-Time Image Processing

Even though the Proc10N is designed for AI and high-bandwidth compute applications, it fully supports Gidel’s image-processing ecosystem for systems that require real-time acquisition and enhancement. All processing is powered by Gidel’s modular real-time preprocessing, enhancement, and advanced compression IPs (.JPEG, Lossless, Quality+).

Inline enhancement features include:

• High Dynamic Range (HDR) – Captures superior details in high-contrast lighting conditions.
• White Balance – Maintains color accuracy across variable lighting conditions.
• Dynamic Luminance Balance – Preserves consistent brightness under changing illumination.
• Gamma Correction – Optimizes brightness and contrast for improved clarity.

Real-time compression with Dynamic ROI delivers:

• Extended recording times without compromising quality.
• Lower transmission bandwidth for efficient data handling.
• Accelerated offline compression, reducing storage needs and post-processing time.

Integrated Vision Architecture for Embedded FPGA Processing

Gidel’s vision architecture, implemented on the Proc10N FPGA modules, supports both the InfiniVision multi-camera vision system and the ProcFG deterministic image acquisition system. Together, they address high-bandwidth acquisition, synchronization, buffering, and scalable processing requirements in demanding vision platforms.

InfiniVision enables flexible, synchronized acquisition across large camera arrays and multi-stream configurations, while ProcFG provides a deterministic acquisition model optimized for fixed timing, guaranteed frame capture, and line-scan–oriented pipelines. This dual approach allows Proc10N module to handle multiple high-speed data sources, merge parallel sensor inputs, and maintain predictable real-time behavior under sustained load.

With high-performance FPGA resources, on-board memory, and deterministic dataflow control, the Proc10N module support stable acquisition and real-time image processing at scale. As a result, they deliver reliable and predictable performance in high-throughput, high-resolution vision systems where bandwidth density and timing control are critical.

Flexible Integration Options

The Proc10N adapts to diverse integration needs. It can operate as a standalone AI/processing engine or as part of a fully customized solution including FPGA-based image enhancement, real-time compression, multi-sensor aggregation, and advanced I/O control. As a result, the Proc10N provides exceptional flexibility and scalability for high-bandwidth vision, AI, and sensing architectures.

SDK, Development Tools, and Integration

The Proc10N is supported by Gidel’s SDK, providing intuitive GUIs and APIs for streamlined system integration. Additionally, the ProcVision Suite offers advanced FPGA programming, debugging, and validation tools. Consequently, developers can rapidly build custom dataflows, AI pipelines, and high-bandwidth processing systems while reducing development time and integration risk.

The module also benefits from Gidel’s advanced multi-port memory architecture, enabling efficient access to HBM2 resources and supporting hundreds of simultaneous operations. Furthermore, each module is supported by Gidel or customer-designed carrier boards, enabling immediate system bring-up and rapid prototyping. As a result, development cycles are shortened, reliability is improved, and time-to-market is accelerated — making the Proc10N ideal for fast-paced, mission-critical industries.

Why Choose the Proc10N? (FPGA for AI module)

• Altera Stratix® 10 NX FPGA with embedded Tensor Blocks.
• HBM2 memory delivering 10× higher DRAM/SRAM bandwidth.
• 143 INT8 TOPS / FP16 TFLOPS for state-of-the-art AI workloads.
• 1,600 Gb/s customizable I/O for high-throughput applications.
• InfiniVision IP for synchronized acquisition from 100+ sensors.
• Compact design suitable for advanced embedded/edge systems.

For smaller FPGA modules, see the FDB series.