The human brain is among the most efficient, sophisticated systems in the known universe, deftly handling pattern and speech recognition, information processing, and even power consumption. Plus, it only weighs about three pounds and fits inside a skull.

"The brain is one of the best templates available for big data analysis and classifications," explains Yang (Cindy) Yi, an ECE assistant professor.

Yi was awarded the National Science Foundation (NSF) Faculty Early Career Development (CAREER) Award to design three-dimensional (3D) neuromorphic integrated circuits—ICs that mimic the human brain. The CAREER grant is one of the NSF's most prestigious awards, given to junior faculty members who are expected to become academic leaders in their fields. "My dream for this technology is that it will improve the quality of human life," says Yi. "We want to build chips to model some of the lost or damaged brain functions, allowing people who have suffered some brain injury to reclaim their former lives or move forward to new ones. "With the grant, Yi will address design challenges, including architecture, integration, speed, and efficiency.

"The human brain evolved to solve a huge number of complicated problems," says Yi. "Which makes our job easier."

Based on architecture that mimics bio-neurological processes, Neuromorphic Computing (NC) systems "leverage evolutionary behaviors to address specific problems that have not been solved by current CMOS (complementary metal-oxide-semiconductor) digital computing," says Yi.

NC systems are poised to surpass 100 million "neurons" with 1 trillion "synaptic connections" within the near future, Yi notes. They will require high complexity, high connectivity, and massively parallel processing to accomplish increasingly demanding computational tasks.

Traditional integration will be incapable of meeting these requirements, but Yi is exploring this technology in a new dimension—literally.

One of the more obvious advantages the human brain holds over current integrated circuit technologies, is that it's not flat. Current 2D integrated circuit technology is approaching its physical and material limits, says Yi. She is investigating how 3D integration technology can be used to create a neuromorphic system that is compatible with current technology—while operating at high system speed with high density and significant parallel processing, low power consumption, and a small design area.

By combining the computational capacity of NC networks with the scalability advantages of 3D integration, Yi's team will be designing NC circuits and systems that more closely emulate the brain's information-processing infrastructure.

Specifically, they will be exploiting time-dependent neural coding and delay-based dynamic nonlinear transfer. This work builds on Yi's extensive research over the past three years, when she fabricated three chips to mimic neural functions.

The first and second chips focused on temporal encoding, one of the main encoding schemes in brain cells. Yi and her team were able to encode the chips with multiple neural codes that operated simultaneously at different speeds while carrying complementary information.

"To the best of our knowledge, the neuron circuit we developed and tested is the first to present sensory data in this way," says Yi.

The third chip incorporates delayed feedback of computing nodes. In this one, Yi is designing the chip to mimic the nonlinear function and agile chaos of the human brain.

To improve reliability and robustness of NC circuits and systems, Yi and her team will be reconfiguring and adapting idle through-silicon vias (TSVs)—high performance electrical connections that run through silicon—as membrane capacitors. "Membrane capacitors typically occupy a significant portion of a chip's design area, and by using idle TSVs to pull double duty, we can substantially reduce design area and boost chip performance," explains Yi.

"If successful, this technology could fuel potentially disruptive capabilities in real-time data analysis, time-series predictions, environmental perception for autonomous operations, and dynamic control systems," says Yi. It could also improve the performance of current and future systems by significantly decreasing power, size, and weight budgets, and by enabling embedded and retrofit applications on legacy, mobile, and remote platforms.

Other applications of Yi's work could improve computing efficiency in wireless communication, cybersecurity, and big data analysis. "Incorporating brain functionality will be a revolutionary change for the field," says Yi. "It's a very exciting time to be involved."