VLSI and Neural Net Systems (VNNS) Laboratory

The VNNS group designs, builds, and tests innovative architectures that span a wide variety of VLSI applications in electrical and biological fields.

Chips designed using digital, analog, and subthreshold methodologies are realized using CMOS, BiCMOS, and Bipolar technologies. Applications include biomimetic sound source localization, neural-net image processing, integrated photonic devices and parallel photonic testing, automatic partial-valued dynamic logic synthesis, single-chip large-molecule and DNA analyzers, and neural tissue interface chips as well as high-voltage Mems-mirror drivers (See Figure 1). The group is equipped with a full suite of design tools and testing instrumentation for analog and digital systems.

Our latest project is an Acoustic Sound Source Localization using Biomimetic Electronics. (See the movie: ADF_Barebone_Fieldtest_2005_06_17.wmv) Our current effort targets the real-time detection and localization of gunshot fire. The biomimetic approach differs significantly from standard digital signal processing methods. The implementation is inspired by the human ear. The VNNS group achieved the successful implementation of such models into custom made hardware (see Figure 2) with outstanding performance features.

Wafer, Figure 1
Figure 1
Custom Hardware, Figure 2

The system can be seen as a high quality audio preprocessor with robust data encoding with high accuracy. Robust data encoding refers to the systems capability to self adapt to environmental noise floor changes and does not need any calibration when re-deployed in other environments. The system has also excellent performance in high reverberant environments such as forests, valleys or urban conditions. The system has very high accuracy which allows to scale the sensor spacing between one inch to e.g. two meters. Recent sensor platforms include an Irobot Packbot with 8" sensor separation (Figure 3a) and a helmet based system (Figure 3b).

Figure 2 

The current research is focused on robust gunshot localization. Other applications include the localization of other impulse sound sources such as mortar fire, artillery and tank fire. The system is also capable of vehicle tracking. Using only a 1 inch sensor, different classes of military vehicles could be tracked up to 500m with an error of +/-5 degree around the track. (see the movies TrackinkM60Video.mov, TrackingM602D.wmv)

iRobot Packbot, Figure 3a Kevlar Helmet, Figure 3b
 Figure 3aFigure 3b
Last Modified: Sunday October 21, 2007