Wideband model of on-chip CMOS interconnects using space-mapping technique

✍ Scribed by Xiaochang Liu; Gaofeng Wang; Jia Liu

• Publisher: John Wiley and Sons
• Year: 2011
• Tongue: English
• Weight: 476 KB
• Volume: 21
• Category: Article
• ISSN: 1096-4290
• DOI: 10.1002/mmce.20534

✦ Synopsis

A new wideband model for on-chip complementary metal-oxide-semiconductor (CMOS) interconnects is developed by virtue of a space-mapping neural network (SMNN) technique. In this approach, two subneural networks are used for improving the reliability and generalization ability of the model. This approach also presents a new methodology for data generation and training of the two neural networks. Two different structures are used for the two subneural networks to address different physical effects. Instead of the S parameters, the admittances of sub-block neural networks are used as optimization targets for training so that different physical effects can be addressed individually. This model is capable of featuring frequency-variant characteristics of radio-frequency interconnects in terms of frequency-independent circuit components with two subneural networks. In comparison with results from rigorous electromagnetic (EM) simulations, this SMNN model can achieve good accuracy with an average error less than 2% up to 40 GHz. Moreover, it has much enhanced learning and generalization capabilities and as fast as equivalent circuit while preserves the accuracy of detailed EM simulations.