Current University Projects

Neuromorphic Computing & Bayesian Networks

Over the course of the last two decades, transistor scaling has largely followed the trend known as Moore’s law which hypothesized the number of transistors per square inch on integrated circuits would double every year. This trend allowed the standard processor architecture developed by John von Neumann to continue to increase in speed, performance, and capability for over a decade. No trend lasts forever though, and in recent years Moore’s law has been revised to suggest the doubling transistors per square inch now occurs every 18 or 24 months. In addition to this slowing trend, the high power dissipation of processors in tight spaces leads to extremely high heat generation which restricts the clock speeds and data throughput of these devices. We notice that while processors continue to evolve, their maximum clock speeds have plateaued restricting the speed at which many computations can be performed.

This restriction on the von Neumann architecture has led to studies on different computing topologies often times basing the work off of our understanding of neural architectures or in other words the architecture of a brain. These neural computing areas of research rely not on the speed of a serial computation path, but instead on the interconnectivity of many different computation blocks or nodes. This form of distributed computing relies on concepts such as message passing to relay information across a grid of nodes. One such approach to this neural computing relies on what is known as belief propagation which, as it sounds, relies on each node in an interconnected grid to derive from its inputs a probabilistic belief as to what the answer might be.

4 Node Dynamic Bayesian Network

To simulate and test these theories in IC design, we utilize the following Cadence tools:

Eye Glass Movement Tracking

Diagnosing eye conditions and problems is often rather difficult given the limited exposure that doctor has with the patent in the exam room. Most of the time for adults, a routine exam is filled with detailed questions upon which the doctor relies heavily on the patent's honest and accurate answers. For children, this is an even more difficult problem that is often fraught with errors. To help solve this problem, we can use specially equipped eye glasses to help track the behavior of children to see if they are they may be experiencing difficulties that they may not have brought forward to their doctor during an eye exam.

Utilizing ultra low power accelerometers and easily removable SD storage, a prototype set of glasses is being developed that will track the movements of the child's head along with detecting when the child is actually using his or her glasses. After using these glasses for several weeks, the child will return to the doctor's office to have the data on the SD card analyzed. From this data, doctors will be better able to understand and diagnose younger patents.

To simulate and test these theories in PCB design, we utilize the following Cadence tools: