NG3 AI Accelerator
Powerful · Compact · Efficient
MOSA-Aligned, COTS-Ready AI from a Proven Partner in Defense Solutions
The Ballard NG3 AI Accelerator combines the cutting-edge Hailo‑8R AI processor with the trusted NG3 platform, delivering powerful computational performance and reliable operation in mission-critical environments.
Building on decades of experience supplying rugged COTS avionics computers to defense primes and integrators, the NG3 AI Accelerator brings AI to a platform you already know you can depend on. Seamlessly blending advanced inference capability with secure, hardened embedded computing, it delivers unprecedented edge processing performance—purpose‑built to meet the heightened data demands of modern aerospace and defense systems while supporting MOSA-aligned, open system architectures.
Compact, Mission-Ready AI for Rugged Platforms
The NG3 AI Accelerator is designed for inference in airborne edge computing environments where power, space, and environmental margins are tight. It delivers practical, on‑platform AI without the overhead of data‑center hardware, making it well suited for UAVs, helicopters, and defense systems that operate in demanding conditions. It combines:
- Small size and low weight - suitable for airborne edge computing and compact avionics bays
- High‑efficiency AI inference - up to 13 TOPS at ~3 TOPS/W for real‑time edge processing
- Rugged - military‑grade design for shock and vibration
- Native avionics interfaces - integrated MIL‑STD‑1553 and ARINC 429/708/717 for direct databus connectivity
- COTS, MOSA‑aligned architecture - simplifying integration and reducing program risk
Four Ways the NG3 AI Accelerator Delivers Real Value at the Airborne Edge
The Ballard NG3 AI Accelerator is engineered to solve the real constraints of aerospace and defense platforms: limited space and power, harsh environments, complex avionics networks, and tight development timelines. By focusing on four core strengths, the NG3 provides a clear path to deploying machine learning and inference at the edge.
Small, Low‑SWaP, Platform‑Optimized Design
The NG3 AI Accelerator fits where traditional servers cannot. Its compact 5.3 x 7.7 x 2.8 in footprint, ~4.5 lb weight, and 28 VDC power input make it easy to integrate into cramped avionics bays, UAVs, and mobile platforms. High‑efficiency AI processing lets you add meaningful compute without major changes to power, cooling, payload, or endurance.
Rugged, Mission‑Ready Reliability
Built on the proven NG3 architecture, the NG3 AI Accelerator is designed to operate from –40 to +55°C with conduction or convection cooling in a sealed, fanless enclosure. Alignment with DO‑160, MIL‑STD‑810, MIL‑STD‑461, and MIL‑STD‑704 means it can handle shock, vibration, EMI, and power variations typical of helicopters, fixed‑wing aircraft, ground vehicles, and unmanned systems.
Integrated Avionics and System I/O
NG3 integrates AI processing and avionics I/O in a single unit. It provides native MIL‑STD‑1553 and ARINC 429/708/717, plus Ethernet, serial, CAN, USB, discretes, DVI, and audio, enabling direct connection to aircraft databuses and sensors without additional gateways. This cuts wiring and weight, reduces latency, and allows NG3 to serve simultaneously as a mission computer, data converter, recorder, and AI engine.
Efficient Edge AI and Developer‑Friendly Software
With up to 13 TOPS at roughly 3 TOPS per watt from the Hailo‑8R NPU, NG3 delivers efficient on‑device inference for vision, signal, and sensor‑fusion workloads. Models from TensorFlow, PyTorch, Keras, TensorFlow Lite, and ONNX can be compiled with Hailo’s tools and run via ONNX and GStreamer. Rocky Linux 9 and the Astronics BTIDriver API provide a familiar environment, easing integration with existing avionics and data acquisition software.
By combining low‑SWaP hardware, rugged construction, integrated avionics I/O, and a modern AI toolchain, the Ballard NG3 AI Accelerator offers a focused, practical path to deploying edge AI in aerospace and defense systems while keeping risk, complexity, and footprint under control.
Ready for Deployment as Your Next COTS Solution
Optimized Performance
- Intel Atom E3950 Quad-Core at 1.6 GHz (2.0 GHz burst)
- 13 TOPS processing power with Hailo-8R
- 3 TOPS per watt, optimized for low-SWaP platforms
Edge AI, Real Missions
The Ballard NG3 AI Accelerator powers next-generation edge AI across mission-critical aerospace and defense platforms, from avionics systems for flight management, sensor fusion, and health monitoring to UAVs for autonomous navigation, object detection, and situational awareness, and electronic warfare systems for real-time threat detection and response.
Cybersecurity Capabilities
In today’s defense landscape, data security is paramount. With hardware-level encryption, secure memory interfaces, and sanitization options, the Ballard NG3 AI Accelerator meets the highest cybersecurity standards, ensuring sensitive mission data is protected.
FAQ
Why Hailo?
Hailo provides the AI technology within the necessary envelope to perform superbly on the NG3 hardware platform. They have demonstrated commitment and stability through the AI revolution with the use of a comprehensive AI application framework that interfaces with many of the standard tools AI developers have come to know and love.
What benefits does the pre-integrated Hailo AI Accelerator offer?
This solution is pre-validated and tested for use in aircraft environments. Specifically, thermal management of the AI module was designed specifically for the NG3 product to ensure reliable and consistent inference performance within the harsh aircraft environment. In addition, you can count on support from Astronics as databus experts and pioneers every step of the way. Examples and whitepapers are provided on top of the already excellent documentation provided by Hailo.
How can I get started with development quickly?
The Ballard NG3 AI Accelerator has the Hailo driver and middleware preloaded and validated on the unit. The NG3 SDK provides a working example that demonstrates real-time AI inference on the device. Customers that wish to evaluate the NG3 solution are welcome to request a free loaner unit. As an alternative, customers already familiar with the NG3 or wishing to evaluate the AI workflow using the same silicon as the Ballard NG3 AI Accelerator can acquire an AI HAT+ that installs easily on top of a Raspberry Pi 5. The recommended configuration for similarity with the NG3 would a Raspberry Pi 5 8GB with an AI HAT+ Hailo-8L variant. This gives a nearly identical software stack and performance capabilities for AI applications as available on the NG3.
How do I collect data from an aircraft for my models?
Astronics provides an SDK for the NG3 that includes drivers and example code for accessing Avionics databusses. Datasets can be built quickly and easily using the BTIDriver API and stored on the ample built-in device storage or sent over a TCP/IP network.
The Versatile NG Family: Compute, Control, Convert
Secure, Pre-validated COTS Solutions
The NG3 AI Accelerator is a member of the NG Family which consists of three distinct series of products of varying capability that share similar design features. At their core, they are all highly-flexible systems that include multi-protocol avionics I/O coupled with a high-throughput Ethernet backbone for inter-device communication and compatibility with a variety of avionics systems.
Deploy Exactly What You Need
Each variant is optimized for different levels of deployment depending on the unique needs of your program. Whether your requirement is for a versatile computer with high amounts of avionics I/O and audio/video capability, a simple converter to move MIL-STD-1553 to Ethernet, or something in between, the NG Family has you covered.
Additional Information
Let's Start the Conversation
For more information or to request a quote, fill out the form below or call us at: +1.425.339.0281
Astronics AES
12950 Willows Road NE
Kirkland, WA 98034 USA
+1.425.339.0281
Ballard.Sales@astronics.com
