DEEPCircuit – Configurable Deep Learning Framework on FPGA

DeepCircuit deploys deep learning applications directly on FPGA fabric without the hurdle of VHDL programming. Upload your trained machine learning model and create code for your FPGA to achieve the lowest latency and power consumption for embedded applications without hassle.

One-Click to Implement Deep Learning on FPGA

Deploy deep learning applications directly on FPGA fabric without the hurdle of VHDL programming. Upload your trained machine learning model and create code for your FPGA to achieve the lowest latency and power consumption for embedded applications.

Simple
DeepCircuit eliminates complexity for end users and enables them to quickly develop and deploy deep learning applications on FPGA.

Efficient
DeepCircuit provides FPGA empowered implementation of deep neural networks, offering low-latency and low-power consumption for a wide range of embedded systems.

Optimized
Deep Circuit instantly provides performance metrics so developers can focus on optimizing accuracy, latency, and power consumption of the system.

Downloads

Features

Simple User Interface

Provides a simple web interface for users to simply upload trained models and test data, and run DeepCircuit to create FPGA code and view performance metrics.

Multiple Models

Supports multiple deep learning algorithms, such as FNN and CNN, from software tools such as TensorFlow, Keras, Matlab, Caffe, and CNTK.

Performance

Provides 8 bit and 16 bit fixed-point precision implementations directly on the FPGA fabric, and compares the performance with floating-point software computations.

FPGA Only Designs

Supports different FPGA platforms from Xilinx and Intel, and eliminates the need for additional computational resources such as CPU and GPU.

Expedite Development

Provides neural network source code directly for your FPGA project reducing coding time.

Use Cases

RF Signal Classification

RF signals (received by RF front ends) are classified quickly and accurately, characterizing the RF spectrum in real-time.

Spectrum Adaptation

Communication systems parameters (e.g., DSA, MIMO, and beamforming) are optimally selected to adapt to spectrum (channel, interference and traffic) dynamics.

Traffic Sign Classification

Traffic signs from images or videos are classified without hand-crafted features, supporting self-driving technologies.

Target Recognition

Targets of interest (e.g., vehicles or structural defects) are detected and classified from overhead imagery (e.g., from UAVs) for surveillance and inspection applications.

Biosensor Data Analysis

Neurological states of individuals are identified from body sensor data (e.g., body temperature, heart rate, arterial oxygen level), supporting healthcare applications.