In embedded and industrial applications, system designers increasingly demand platforms that combine compactness, AI capability, and resilience against harsh conditions. The PCIe104-Jet rises to this challenge as a rugged, PCIe104-form-factor carrier board that brings NVIDIA® Jetson™ compute power to the AI-on-the-Edge, with full compatibility for modular, high-performance designs in challenging environments.
Built for Rugged AI Applications with PCIe104
The PCIe104 standard is renowned for enabling stackable, high-speed PCIe systems in compact and mechanically robust form factors. Designed without backplanes or card cages, PCIe104 boards are ideal for applications exposed to vibration, extreme temperatures, and EMI, such as defense, transportation, robotics, and industrial automation.
The PCIe104-Jet leverages this proven architecture, offering PCIe/104 Type 1 and Type 2 connectors for expansion. In particular, the board supports:
-
Stack-up mode (Type 2 connector): Full compatibility with standard PCIe104 peripheral modules, enabling system expansion with PCIe x1 devices and USB 3.0.
-
Stack-down mode (Type 1 connector): Exclusively supports SundanceDSP FPGA modules, including the PCIe104z and PCIe104-RFSoC, allowing the integration of high-speed logic and real-time signal processing.
This selective compatibility ensures tight integration and software support between Jetson GPUs and Sundance’s powerful FPGAs, creating a streamlined pipeline for hybrid computing at the edge.
Key Features
-
Supports NVIDIA Jetson Modules: Nano, Orin NX, Orin Nano, TX2 NX, Xavier NX, and Orin NX MAXN_SUPER 40W in a 260-pin SODIMM format
-
PCIe/104 Type 2 Stack-Up Connector:
-
Up to 4x PCIe x1 lanes and 2x USB 3.0 ports (via assembly option)
-
Fully compatible with standard PCIe104 peripherals
-
-
PCIe/104 Type 1 Stack-Down Connector:
-
PCIe x4 (x2 for TX2 NX)
-
Supports only Sundance DSP PCIe104 FPGA boards (PCIe104z, PCIe104-RFSoC)
-
-
Dual 22-pin CSI camera interfaces, Raspberry Pi compatible
-
Gigabit Ethernet via Harting iX industrial connector
-
HDMI output, UART via USB-C
-
Optional microSD, SPI, I2C, and GPIO headers (based on Jetson module used)
-
User and stack error LEDs
Block Diagram Overview
This block diagram illustrates the layout and interconnections of the PCIe104-Jet board:
Central Processing
At the center is the 260-pin SODIMM connector, which accepts a variety of Jetson modules. These modules interface with all onboard I/O and PCIe links.
Stack Interfaces
-
SIDE 1 (Top, Stack-Up Mode):
-
PCIe Type 2 connector provides up to 4x PCIe x1 lanes and 2x USB 3.0 ports, depending on module configuration.
-
Compatible with standard PCIe104 add-on boards (e.g., data acquisition, storage, networking).
-
-
SIDE 2 (Bottom, Stack-Down Mode):
-
PCIe Type 1 connector offers PCIe x4 (x2 for TX2 NX).
-
Only compatible with SundanceDSP boards, such as the PCIe104z and PCIe104-RFSoC, to ensure tight integration for hybrid computing.
-
Camera and I/O Interfaces
-
Two Raspberry Pi-compatible 22-pin camera connectors for stereo or dual-camera applications.
-
HDMI output for display, connected via a level-shifted DP1 signal.
-
USB-C (CP2105 bridge) for UART0 and UART1 communication.
-
1Gb Ethernet via iX industrial connector.
-
Optional microSD, SPI, I2C, and GPIO headers (available on select Jetson modules).
-
Additional USB2 and 12C1 lines routed to an onboard mPCIe connector.
Power and Control
-
Power is supplied via an external PWR connector.
-
Onboard power regulation and MUX circuitry manage voltage rails (5V, 12V, 3.3V) and signal selection.
-
User LED and stack error LED for diagnostics.
Applications
The PCIe104-Jet, combined with Sundance DSP’s PCIe104z or PCIe104-RFSoC FPGA boards in stack-down mode, forms a powerful hybrid architecture. In these applications, the FPGA provides ultra-low-latency, deterministic pre-processing or high-speed signal acquisition, while the Jetson GPU handles AI inference, complex decision-making, and data post-processing.
1. AI-Enhanced Software-Defined Radio (SDR)
Use Case: Spectrum monitoring, cognitive radio, radar, and electronic warfare.
-
FPGA Role: The FPGA handles real-time high-speed digitization of RF signals (via ADCs), digital down-conversion, FFTs, and protocol-specific framing/deframing.
-
GPU Role: The GPU performs AI inference for tasks like signal classification, anomaly detection, modulation recognition, and adaptive decision-making.
-
Why Both? RF data must be processed deterministically and at line rate, achievable only with FPGAs. AI models applied to spectrograms or baseband data can identify threats or patterns that traditional DSP alone cannot.
-
Example: A cognitive radio system that dynamically avoids interference by analyzing the spectrum in real time and using AI to choose optimal communication parameters.
2. Real-Time Industrial Vision Systems
Use Case: High-speed visual inspection, defect detection, and smart manufacturing.
-
FPGA Role: Interfaces with high-frame-rate MIPI cameras; implements frame buffering, custom triggering, synchronization, and ROI extraction.
-
GPU Role: Performs deep learning inference for image segmentation, object detection, or classification (e.g., identifying surface defects).
-
Why Both? The FPGA offloads camera interface and real-time signal conditioning, ensuring minimal CPU/GPU loading, while the GPU enables sophisticated visual recognition using CNNs or transformer-based models.
-
Example: An inline inspection system that captures 500+ FPS images of a production line, flags defects using AI, and triggers actuators via FPGA-controlled outputs.
3. Intelligent Radar and LIDAR Processing
Use Case: Obstacle detection, autonomous navigation, perimeter security.
-
FPGA Role: Digitizes and processes raw radar or LIDAR signals (e.g., beamforming, range-Doppler computation).
-
GPU Role: Interprets point clouds or range-Doppler maps using AI for object classification, tracking, or environmental understanding.
-
Why Both? FPGA handles high-rate sensor data with deterministic latency, while GPU enables real-time semantic interpretation, which is computationally intensive.
-
Example: A perimeter surveillance system that fuses radar and visual input to detect and classify intrusions in all weather conditions.
4. Embedded Sensor Fusion for Robotics and UAVs
Use Case: SLAM (Simultaneous Localization and Mapping), autonomous navigation, multi-modal sensing.
-
FPGA Role: Fuses data from inertial sensors (IMUs), rotary encoders, and time-of-flight sensors; ensures tight timing correlation.
-
GPU Role: Performs visual odometry, depth estimation, SLAM graph optimization, and path planning using AI models.
-
Why Both? FPGA ensures real-time fusion and deterministic control loop timing; GPU handles complex spatial reasoning and mapping tasks that benefit from parallel computation.
-
Example: An autonomous drone that integrates IMU, GPS, stereo vision, and radar to map and navigate an unknown environment.
5. AI-Driven Acoustic Signal Analysis
Use Case: Fault detection in machinery, structural health monitoring, and underwater acoustic processing.
-
FPGA Role: Captures high-bandwidth microphone or hydrophone data; performs FFT and filtering at high speed.
-
GPU Role: Classifies acoustic signatures using trained neural networks (e.g., identifying mechanical anomalies or underwater targets).
-
Why Both? FPGA enables high-throughput acquisition and low-latency DSP, while GPU identifies complex or subtle patterns not detectable through thresholding or rule-based approaches.
-
Example: A predictive maintenance system that detects bearing wear or cavitation in motors based on learned audio signatures.
Software Ecosystem
In addition to support for NVIDIA’s Jetson SDKs, SundanceDSP provides a tailored Board Support Package (BSP) for integration with its FPGA hardware. This BSP includes device drivers, middleware, and reference designs to accelerate development and deployment.
Conclusion
The PCIe104-Jet delivers the perfect balance of edge AI computing, industrial-grade robustness, and modular expansion. Its stack-up support for standard PCIe104 peripherals and stack-down support exclusively for Sundance DSP FPGA boards ensures high-performance integration of GPU, CPU, and FPGA in a compact, rugged system.
Whether for machine vision, automation, robotics, or AI-enhanced RF applications, the PCIe104-Jet provides the scalable foundation to power next-generation embedded solutions — even in the most demanding environments.