Research in LA.P.I.S.


Our research is focused on the general area of computer engineering with specific interests in:


Multiprocessor Systems

In the area of Multiprocessor Systems and Interconnection Networks, we are investigating generalized interconnects, termed Hypercycles, which allow flexibility in the structuring of a concurrent computer system in both the number of nodes and the capacity of the interconnect.

In relation to these networks, we have devised a number of routing strategies including oblivious and non-oblivious ones and proved deadlock-free properties.

We have devised an allocation strategy and proved that it is statically optimal, and designed and fabricated (in VLSI or Gate Arrays) a number of components that implement some of the proposed routing strategies.

We have developed a specialized Hardware Description Language (CoDeL), used to express the routing algorithm, and producing synthesizable VHDL code.

Additionally, we have developed optimal routing algorithms for specific problems (e.g. total exchange) on networks having topologies such as rings, trees and fat trees.


Neural Networks

In the area of Neural Networks, we have investigated the structure of certain types of recurrent neural networks and proven that under certain conditions they are asymptotically stable.

We have devised learning laws for these networks, and have used them to model nonlinear dynamical systems.

We are currently investigating improvements in the learning laws so that we can effect speedier convergence.

We are using this type of neural networks in fault detection and diagnosis.


Synthesis Automation

In the area of Synthesis Automation, we have developed techniques that validate the timing of an interface design before its implementation.

Our techniques are based on the timing analysis of protocols involved and obtain tight bounds on interface path delays.

We are currently refining our techniques to use statistics on the timing information. This information can lead to the derivation of reliability measures for the design instead of just bounds on the interface-path delays.


Fault Detection and Diagnosis of Large Communication Networks

In collaboration with the ROGERS Canadian Cable Labs Fund, we are developing an automated diagnosis environment for Cable Television Distribution Networks.

Neural Network and Expert Systems techniques are utilized to provide accurate and real time fault detection and diagnosis.

Last updated: April, 1996