Home ] Up ] LS Architectures ] PWL Functions ] Sensor Networks ] Nanotech ] Signal Processing ]


Research

 

The amazing progress in microelectronics circuit integration density  gives the possibility of integrating massive arrays of devices. Moreover, the advent of nanotechnology will assure the current rate of density increase in the following decades. Therefore, it becomes a necessity to develop  large scale computing architectures. In this specific subject, I am working on parallel computing paradigms that include Cellular Automata and Cellular Neural Networks (CNN). My current research focuses on CMOS efficient implementation of CNN for imaging applications. This includes mainly the so-called Simplicial-based CNN (S-CNN).  We have developed a first prototype in 0.5um technology; future research includes higher density designs using more sophisticated technologies.   Image Gallery of the S-CNN

The objective of this project is the design of a full- system low-power integrated circuit to localize acoustic targerts.We have designed several prototypes in different technologies. We have shown that an accuracy of 1 degree can be reached with 15uA at 3V. 

Sensor Networks for Local Applications 

Sensor Networks provide a way to measure an environmente, whether natural, urban or industrial, to collect information and take decisions upon them. We are developing low-power interfaces, sensors and systems for a variety of applications that are of particular interest in Argentina, for example: urban acoustic contamination, cargo tracking, early fire alarms in forests, security and surveillance in farms and closed  scenarios, precise agriculture, etc. 

Nonlinearities are present in the majority of real life applications, and they often pose a serious challenge to design engineers. Techniques for the treatment of linear systems are widely known and have been widely studied. In contrast, techniques for nonlinear systems are not that well developed and only apply to particular cases. Piecewise-Linear functions provide a way to accurately approximate a nonlinear function with a collection of linear functions, each of them valid in a particular region of the domain space. In this way, techniques originally designed for linear systems can be extended to nonlinear systems. My interests on PWL functions are related to the theory of canonical representations,  numerical methods for the approximation of nonlinear functions and high-speed CMOS circuits for nonlinear control and processing.

This year, UMC has already delivered its 90nm CMOS process (gate lenght app. 50nm, 1V) and  TSMC will do the same in next months.  Intel is actually producing 70nm gate length transistors commercially, and has prototypes of 15nm gate length under testing. This reduction trend in CMOS technology is expected to continue roughly until 2016, according to the predictions of the International Technology Roadmap for Semiconductors. Due to this, a whole spectra of emerging devices and technologies are being studied.I have been exploring implementations of the S-CNN using Resonant Tunneling diodes (RTD) devices and memories.