论文信息 - A 4-transistor Euclidean distance cell for analog classifiers

A 4-transistor Euclidean distance cell for analog classifiers

The areal efficiency of a high-throughput parallel classifier depends strongly on the area of the basic distance cell because classifier architectures contain a large array of such cells. For this reason, reducing the circuit complexity of a distance cell is expected to have a magnified impact on the overall size. A new squared Euclidean distance-cell configuration comprising four MOSFETs and two capacitors is proposed. The cell is based on symmetrical capacitive template storage and current-domain readout, which help reduce size while increasing speed and precision. A typical cell design in a mainstream 0.8 /spl mu/m CMOS technology occupies 17.2/spl times/21.9 (/spl mu/m/sup 2/) silicon area representing an areal density of 2650 cells per square millimeter. The article includes a description and analysis of the circuit, identifies design constraints and presents an error analysis.

U. Cilingiroglu | D.Y. Aksin

[1] P. Heim,et al. Precise analogue synapse for Kohonen feature maps , 1993, ESSCIRC '93: Nineteenth European Solid-State Circuits Conference.

[2] Michel Verleysen,et al. Analog implementation of a Kohonen map with on-chip learning , 1993, IEEE Trans. Neural Networks.

[3] K. R. Lakshmikumar,et al. Characterisation and modeling of mismatch in MOS transistors for precision analog design , 1986 .

[4] Oscal T.-C. Chen,et al. A VLSI neural processor for image data compression using self-organization networks , 1992, IEEE Trans. Neural Networks.

[5] J.S. Pena-Finol,et al. A MOS four-quadrant analog multiplier using the quarter-square technique , 1987 .

[6] G. Cauwenberghs,et al. A low-power CMOS analog vector quantizer , 1997 .

[7] Yuping He,et al. A charge-based on-chip adaptation Kohonen neural network , 1993, IEEE Trans. Neural Networks.