Hardware Implementation of Kohonen Neural Network using CMOS Circuits
Abstract
In this paper efficient hardware implementation of Kohonen Neural Network (KNN) has been discussed. To achieve this, current mode has been chosen as the basis of circuit level implementations. By considering this, a novel low power current squarer circuit is introduced which has successfully been employed in the body of the network. Also, the evaluative operations have been carried out using current comparators. Moreover, careful design considerations are utilized to optimize the performance of the proposed architecture. In summary, the main building blocks of proposed architectures are Euclidean Distance Calculation circuit (EDC), Winner Takes All (WTA) circuit and Adaptive Weight Change (AWC) mechanism. The designed process has been done in analog domain to achieve more precision. As a consequence, the notable advantage of this network is its low power consumption and chip area compared to similar digital networks. Consisting of four output neurons each of which has three weights, the circuit is implemented in CMOS 0.18µm process and occupies a chip area equal to0.04 and consumes 0.5mw of power for 1.8V supply. System level and circuit level simulation results have been provided to show the correct behavior of implemented network where the conformity between the results is in the moderate level.
References
[2] Ewaryst Tkacz, Pawel Kostka, Krzysztof Jonderko, and Barbara Mika, “Supervised and unsupervised learning systems as a part of hybrid structures applied in ECG signals classifiers,” Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2005.
[3] Michael Hentrich, “Methodology and Coronary Artery Disease Cure,” SSRN, 2015.
[4] J. Zurada, “Introduction to Artificial Neural Systems”, West Publishing Company, USA, 1992.
[5] S Azizian, K Fathi, B Mashoufi, F Derogarian, “Implementation of a programmable neuron in 0.35μm CMOS process for multi-layer ANN applications,” EUROCON-International Conference on Computer as a Tool (EUROCON), pp. 1-4, 2011.
[6] Golnar Khodabandehloo, Mitra Mirhassani, and Majid Ahmadi, “Analog Implementation of a Novel Resistive-Type Sigmoidal Neuron,” IEEE Transactions on Very Large Scale Integration (VLSI) Systems, Vol. 20, No. 4, April 2012.
[7] Jafar Shamsi, Amirali Amirsoleimani, Sattar Mirzakuchaki, Arash Ahmade, Shahpour Alirezaee, Majid Ahmadi, “Hyperbolic Tangent Passive Resistive-Type Neuron,” 2015 IEEE International Symposium on Circuits and Systems (ISCAS), pp. 581-584.
[8] Sarkis Azizian, Viggen Aziz Aghchegala, Sarhad Azizian, and Mehdi Sefidgar Dilmaghani, “CMOS Bulk-Controlled Fully Programmable Neuron for Artificial Neural Networks,” IETE Journal of Research, Vol. 65, No. 3, pp. 320-328, 2019.
[9] B. Hochet, V. Peiris, G. Corbaz, and M. Declercq, “Implementation of a neuron dedicated to Kohonen maps with learning capabilities,” IEEE Proceedings of the Custom Integrated Circuits Conference, 1990.
[10] V. Peiris, B. Hochet, and M. Declercq, “Implementation of a fully parallel Kohonen map: a mixed analog digital approach,” Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94), 1994.
[11] Tomasz Talaska, and Rafal Dlugosz, “Current Mode Euclidean Distance Calculation Circuit for Kohonen's Neural Network Implemented in CMOS 0.18µm Technology,” Canadian Conference on Electrical and Computer Engineering, 2007.
[12] Rafal Dlugosz, Marta Kolasa, Witold Pedrycz, and Micha Szulc, “Parallel Programmable Asynchronous Neighborhood Mechanism for Kohonen SOM Implemented in CMOS Technology,” IEEE Transactions on Neural Networks, Vol. 22, No. 12, pp. 2091-2104, 2011.
[13] Deboeck, Guido, and Teuvo Kohonen, “Visual explorations in finance: with self-organizing maps”, Springer Science & Business Media, Germany, 2013.
[14] Behzad Ghanavati, and Enayatollah Taghavi Moghaddam. “Low-Voltage CMOS Multiplier Circuit Based on the Translinear Principle,”.Universal Journal of Electrical and Electronic Engineering, Vol. 2, No. 3, pp. 124-127, 2014.
[15] T. Talaska, R. Wojtyna, and Rafał Długosz, “Hardware implemented neural network model with unsupervised learning on silicon,” Int. Workshop MIXDES’2005, 2005.
[16] Tuevo Kohonen, “Self-organizing maps”, Springer Series in Information Sciences, Germany, 2001.
[17] Rafal Dlugosz, Tomasz Talaska, and Witold Pedrycz, “Current-mode analog adaptive mechanism for ultra-low-power neural networks,” IEEE Transactions on Circuits and Systems II: Express Briefs,Vol. 58, No. 1, pp. 31-35, 2011.
[18] Ernesto Araujo, Cassiano R. Silva, and Daniel J. B. S. Sampaio, “Video target tracking by using competitive neural networks,” WSEAS Transactions on Signal Processing, Vol. 4, No. 8, pp. 420-431, August 2008.
[19] Arnold F. Shapiro, and R. Paul Gorman, “Implementing adaptive nonlinear models,” Insurance: Mathematics and Economics, Vol. 26, pp. 289-307, 2000.
[20] R. Dlugosz, T. Talaska, J. Dalecki, and R. Wojtyna, “Experimental Kohonen neural network implemented in CMOS 0.18μm technology,” 15th International Conference on Mixed Design of Integrated Circuits and Systems, 2008.
[21] Alice Wang, Benton Highsmith Calhoun, and Anantha P. Chandrakasan, “Sub-threshold design for ultra low-power systems”, Springer-Verlag, US, 2006.
[22] Amir Fathi, Sarkis Azizian, and Nastaran Sharifan. “Sensors and Amplifiers: Sensor Output Signal Amplification Systems,” In Handbook of Research on Nanoelectronic Sensor Modeling and Applications, pp. 423-504. IGI Global, 2017.
