/❋the FIRST real DPU for all FPGAs❋/

Engineered for the most demanding applications, NeuralBoost delivers failsafe, energy-efficient, and adaptable solutions for next-generation systems.

What Sets Neuralboost Apart?

Easy enabler of AI inference without model restrictions
No add-on or private toolchains, flow based on open-source toolchains TensorFlow and PyTorch
Why does it matter? It runs without restrictions on any FPGA or FPGA SoC device

/❋application examples❋/

Smartlock

Face Detection & Recognition

Check out our Smartlock product here

AI Pilots

Object Detection, Object Tracking, Segmentation & State Estimation

Ultrasound

Segmentation & Semantic Segmentation

Virtual production

Person / Face / Object Tracking

ROBOT PICK & PLACE

3D Object Detection & Pose Estimation

AUTONOMOUS AGENTS

Deep Stereo Vision, Instance Segmentation & Loop Closure

/❋compatibility❋/

AMD Zynq™ 7000

AMD Zynq™ UltraScale™

AMD Artix™/Spartan™ UltraScale™

AMD Virtex™ UltraScale™

AMD Versal™ AI Core/AI Edge

AMD Versal™ Prime/Premium

Microchip PolarFire™/PolarFire™ SoC

/❋support❋/

Open-source support

Custom operating system

Real-time/Determinism

Only FPGA devices

Bare-Metal

Resource configuration

No hardware recompilation

Multiple models

NeuralBoost has the additional following features:

It has a lightweight set C/C++ APIs which are compatible with custom integration from customer in their software environment
It has APIs to control the IP from AMD MicroBlaze™, enabling also devices with no ARM cores (MicroBlaze™ V coming 2025)
It has a fine-grained system for re-compiling targeting different hardware resources
Non-supported layers from the hardware accelerator are scheduled automatically on the ARM NEON accelerator (if ARM core is available)

/❋Supported Neural Networks❋/

CNN

RNN

Transformer

Why NeuralBoost Is Essential for Your Operations

How NeuralBoost Outperforms Competitors

Many systems like Vitis AI struggle with outdated model support and complex toolchains. NeuralBoost offers a straightforward, dependable solution without these limitations.

Energy Efficiency: Operates at 0.5W to 75W, compared to competitor GPUs consuming up to 1200W, reducing energy costs by 40% annually.

Exceptional Durability: Fault-tolerant design ensures operation for up to 30 years, even in harsh environments.

Cost-Effective Updates: Hardware flexibility and over-the-air (OTA) updates eliminate the need for frequent replacements, saving up to 20% on lifecycle costs.

AI APPLICATION SUPPORT

Custom dataset creation
AI training and model creation
Model optimization
Application deployment

FPGA AND EMBEDDED DEVELOPMENT SUPPORT

Integration of the IP in the design
Integration of the low-level code in the application
IP performance optimization
Custom accelerator implementation

/❋volume based fee❋/

Prices ranges depending on device family, number of licenses and end market
The device families are divided as follows:
- 7 series/Zynq™ 7000
- UltraScale™/UltraScale+™ devices
- Versal™ devices
End markets are:
- Consumer
- Industrial/Robotics/Pro AV
- Medical/Automotive
- Aerospace & Defense

Discover how Neuralboost can transform your business.

Please fill out the form to get all the information, we will be happy to deepen your knowledge of features, applications and possibilities of Neuralboost.