One major thrust of Jianhua Joshua Yang’s research in the Electrical and Computer Engineering Department is creating new kinds of memristor devices that, among other uses, can take humans into the so-called “last frontier” of computing: a computer that works like the human brain. No wonder that during his first year as a professor at the University of Massachusetts Amherst, Yang has been awarded three impressive research grants totaling nearly $1.5 million.
The three grants include: $1,095,144 for phase I (an extra $500,000 is pending for phase II) from the Air Force Research Lab (AFRL) for a project entitled “Memristor Crossbar Arrays For Analog and Neuromorphic Computing”; $249,983 from the Intelligence Advanced Research Projects Activity (IARPA) for research on an “Analog Memristor Based Hybrid Computation Engine”; and $99,994 from the Hewlett-Packard Company for “Fundamental Material Research for Unconventional Computing.”
Yang is the director of The Ionic and Electronic Device and Materials (IEDM) Group in ECE. All three of Yang’s recent grants are based on his research studying memristors, or nanoscale devices whose resistance depends on the history of the current/voltage applied to them. In other words, a memristor's present resistance depends on how much electric charge has flowed through it in the past and in which direction. The device remembers its history, which creates its so-called “non-volatility” property (when the electric power supply is turned off, the memristor remembers its most recent resistance until it’s switched on again).
Yang’s well-funded research for the AFRL involves using a memristor crossbar array for neuromorphic computing. The ultimate goal is conquering the last computing frontier, i.e., building a human-brain-like computer.
“The brain remains one of the greatest scientific mysteries,” explains Yang. “Yet, this is not the biggest reason we desperately want to understand and emulate it. The traditional computing technology has revolutionized the world, but it is reaching its fundamental limit. A completely new concept of computing is needed.”
The White House announced in October of 2015 a grand challenge to develop transformational computing capabilities by combining innovations in multiple scientific disciplines. The brain initiative is one of three overarching initiatives in this White House challenge.
As Yang explains, “The superiority of the human brain originates from its different computing paradigm from our digital computers. Our brains process and store information relying on two types of building blocks, 1011 neurons and 1014~15 synapses, which are analog, parallel, adaptive (hardware rather than software learning), and multifunctional devices.”
Circuits based on the traditional complementary metal-oxide semiconductor (CMOS) devices – meaning the semiconductor technology used in the transistors that are manufactured into most of today's computer microchips – have been built to emulate human neurons and synapses at the circuit function level. But we need roughly a million state-of-the-art CMOS chips to emulate a human brain, which is obviously not practical or efficient.
“Therefore,” says Yang, “we need to use devices that share similar physics with neurons and synapses and thus emulate them at the device mechanism level. It turns out that memristors are the nearly ideal devices for such purposes. It has been demonstrated that it takes only one and two memristors to emulate a synapse and a neuron, respectively. Moreover, memristors can be as small as 5nm, much smaller than a real synapse or neuron. The number of memristors required to emulate a human brain can thus fit in an A4 paper.”
As a result, two-terminal conductor/insulator/conductor memristors have drawn great attention in the past decade because these devices hold the promise to realize the two key goals for novel computing. They can potentially lead to highly dense non-volatile random access memories to transform the traditional computing technology and also synaptic electronics to enable the neuromorphic computing architecture mimicking the human brain.
Meanwhile, Yang is doing research for the IARPA using a memristor crossbar array for analog computing, which can be integrated with a traditional digital computer to form a hybrid computation engine. In this hybrid computer, the analog computing based on memristors can efficiently (with a 100-1000x faster speed) handle the tasks that are difficult for the traditional digital computer, such as vector-matrix multiplication, which is important for many applications, such as pattern recognition.
For Hewlett-Packard, Yang is doing fundamental material research to improve memristors for non-volatile random access memories, which may lead to a universal memory that is dense, fast, and nonvolatile. This kind of memory can significantly improve the traditional digital computing technology.
Yang also serves as the co-editor of Applied Physics A (Materials Science & Processing), and he has been chosen as a guest editor for a Nanotechnology special issue on “Non-volatile memory based on nanostructures” in 2011, for an Applied Physics A special Issue on “Memristive and resistive devices and systems” in 2011, and for an IEEE JETCAS special issue on “Solid-state Memristive Devices and Systems” in 2015. (December 2015)