Deep neural networks (DNN) have attracted resurgent interest because of their great potential in implementing artificial intelligence (AI).
One example for the excellent computing power of DNNs is the historic victories of AlphaGo over the best Go players in the world. However, the power consumption of the hardware behind AlphaGo
is extremely high as compared to that of a human brain (~25 W), not to mention that the former has much heavier weight and a much bigger size.
To address the size, weight and power consumption (SWaP) issues, we believe computing systems built upon emergin devices and nanotechnology, for instance memristors, offer an attractive solution. A memristor, also known as a resistance switch or RRAM, is an electronic device whose internal resistance state is dependent on the history of the current and/or voltage it has experienced. Using the conductance state to represent synaptic weight,
non-volatile memristors do not need external power to maintain the states. The programming (training) needs a small amount of energy (<3 fJ) as the device can be extremely small (<2 nm) and the switching speed can be very fast (<1ns). Organized into large arrays or stacked into 3D,
memristors are capable of in-memory computing using physical laws, such as Ohm's law and Kirchhoff's current law to implement vector matrix multiplication (VMM). The current readout at all columns (inference) can be accomplished simultaneously regardless of the array size, leading to massive parallelism in computing.
The new computing paradigm avoids the time and energy needed to access system memory in a digital system (so called 'von Neumann bottleneck').
Furthermore, the physical computing is analog in nature and the network could interface with analog data acquired directly from sensors, reducing the energy overhead from analog-to-digital conversion.
Our goal is to build an energy-efficient hardware platform using emerging electronic devices. Currently our lab is focused on the following main topics: 1) Energy-efficient hardware systems for machine intelligence, security, sensing and communication; 2)
Emerging nanoelectronic devices: design, characterization and understanding; and 3) Enabling fabrication and three-dimensional heterogeneous integration technologies.
We are well-equipped with nanofabrication, electrical measurement and physical characterization facilities for our research. Some of them are owned by our group, some are shared facilities on
campus that we have full access to. A list of most relevant equipment is on the next page.
We truly appreciate the generosity and foresight of all our sponsors!