Jump to page content

C-BRIC

This center is working to propel cognitive computing, driven by advances in machine learning theory and applications.

The Center for Brain-inspired Computing Enabling Autonomous Intelligence (C-BRIC) is a five-year, Purdue-led center supported by $32 million from the Semiconductor Research Corp. (SRC) via its Joint University Microelectronics Program (JUMP). JUMP provides funding from a consortium of industrial sponsors and the Defense Advanced Research Projects Agency (DARPA). The SRC operates research programs in the U.S. and globally that connect industry to university researchers, deliver early results to enable technological advances, and prepare a highly trained workforce for the semiconductor industry. The Indiana Economic Development Corporation also supports C-BRIC with $1.5 million for an intelligent autonomous systems laboratory at Purdue.

The past decade has seen tremendous progress in cognitive computing, driven by advances in machine learning (ML) theory and applications. Cognitive systems are poised to transform various industries and society at large, spurring demand for computing and semiconductor products. The use of ML for various analytics tasks, including web searches, object recognition in images and videos, and speech recognition, is only the first wave of cognitive computing.

The next wave will usher artificial intelligence (AI) into our lives pervasively through autonomous intelligent systems, such as self-flying drones, self-driving automobiles, and interactive personal robots. It will provide significant capabilities to the nation’s economic robustness and military readiness.

Established in 2018 as a multi-university center, C-BRIC is working to deliver key advances in cognitive computing, enabling a new generation of autonomous intelligent systems. C-BRIC is co-developing neuro-inspired algorithms and theory, neuromorphic computing fabrics and distributed intelligence to address industry challenges. The research team’s expertise spans ML, computational neuroscience, theoretical computer science, neuromorphic hardware, distributed computing, robotics, and autonomous systems.

C-BRIC is moving closer to emulating what the brain can do — sometimes better and faster than the brain. However, hardware cost is still rising. Inefficiencies come from not having the right algorithm, hardware architectures, and circuits and devices. Purdue’s C-BRIC researchers are continuing to work to enable autonomous intelligent systems by improving compute efficiency and robustness of cognitive tasks through cross-layer innovations from algorithms to hardware.

Indicating its reach and value, C-BRIC has graduated more than 70 PhD students. Graduates have gone on to positions with such industry partners such as ARM, IBM, Intel, Micron and Samsung, and many have accepted tenure-track faculty positions at highly ranked universities.

Achievements by C-BRIC Researchers at Purdue Include These Innovations:

PUMA
(Programmable Ultra-efficient Memristor-based Accelerator for Machine Learning Inference)

The simulator enables the acceleration of a wide variety of ML inference workloads, showing more than a 100x energy improvement compared with standard GPUs. PUMA helps to better estimate such system-level metrics as energy consumption and onboard data latency.

RxNN
(Crossbars Neural Networks) and GenieX (Generalized approach to modeling non-idealities in crossbars)

These simulators evaluate crossbar-based architectures for next-generation AI hardware. The tools have been shared with Samsung, Intel and IBM for internal use. GenieX is also made available to the broader research community through the Github.

AxIS
(Approximate Inference System)

Using a smart camera system, AxIS performs synergistic approximations across different subsystems in a DNN-based inference system, leading to energy benefits in image classification and object detection.

DIET-SNN
(Direct Input Encoding with Leakage and Threshold Optimization in Deep Spiking Neural Networks)

To help provide a good way to train large-scale SNNs, C-BRIC has pioneered research on converting deep learning networks to SNNs for lower latency systems. The DIET-SNN operates with asynchronous spikes distributed over time to allow greater computational efficiency on event-driven hardware.

IMPULSE
(Digital In-Memory Processing Based Low-Power Spiking Neural Inference Engine)

A chip capable of performing brain-like sequential learning tasks. IMPULSE is an SNN accelerator used to achieve real-time data and greater energy efficiency.

Related Link: https://cbric.org