Reliable chip design from low powered unreliable components

The pace of technological improvement of the semiconductor market is driven by Moore’s Law, enabling chip transistor density to double every two years. The transistors would continue to decline in cost and size but increase in power. The continuous transistor scaling and extremely lower power constraints in modern Very Large Scale Integrated(VLSI) chips can potentially supersede the benefits of the technology shrinking due to reliability issues. As VLSI technology scales into nanoscale regime, fundamental physical limits are approached, and higher levels of variability, performance degradation, and higher rates of manufacturing defects are experienced. Soft errors, which traditionally affected only the memories, are now also resulting in logic circuit reliability degradation. A solution to these limitations is to integrate reliability assessment techniques into the Integrated Circuit(IC) design flow. This thesis investigates four aspects of reliability driven circuit design: a)Reliability estimation; b) Reliability optimization; c) Fault-tolerant techniques, and d) Delay degradation analysis. To guide the reliability driven synthesis and optimization of combinational circuits, highly accurate probability based reliability estimation methodology christened Conditional Probabilistic Error Propagation(CPEP) algorithm is developed to compute the impact of gate failures on the circuit output. CPEP guides the proposed rewriting based logic optimization algorithm employing local transformations. The main idea behind this methodology is to replace parts of the circuit with functionally equivalent but more reliable counterparts chosen from a precomputed subset of Negation-Permutation-Negation(NPN) classes of 4-variable functions. Cut enumeration and Boolean matching driven by reliability-aware optimization algorithm are used to identify the best possible replacement candidates. Experiments on a set of MCNC benchmark circuits and 8051 functional microcontroller units indicate that the proposed framework can achieve up to 75% reduction of output error probability. On average, about 14% SER reduction is obtained at the expense of very low area overhead of 6.57% that results in 13.52% higher power consumption. The next contribution of the research describes a novel methodology to design fault tolerant circuitry by employing the error correction codes known as Codeword Prediction Encoder(CPE). Traditional fault tolerant techniques analyze the circuit reliability issue from a static point of view neglecting the dynamic errors. In the context of communication and storage, the study of novel methods for reliable data transmission under unreliable hardware is an increasing priority. The idea of CPE is adapted from the field of forward error correction for telecommunications focusing on both encoding aspects and error correction capabilities. The proposed Augmented Encoding solution consists of computing an augmented codeword that contains both the codeword to be transmitted on the channel and extra parity bits. A Computer Aided Development(CAD) framework known as CPE simulator is developed providing a unified platform that comprises a novel encoder and fault tolerant LDPC decoders. Experiments on a set of encoders with different coding rates and different decoders indicate that the proposed framework can correct all errors under specific scenarios. On average, about 1000 times improvement in Soft Error Rate(SER) reduction is achieved. Last part of the research is the Inverse Gaussian Distribution(IGD) based delay model applicable to both combinational and sequential elements for sub-powered circuits. The Probability Density Function(PDF) based delay model accurately captures the delay behavior of all the basic gates in the library database. The IGD model employs these necessary parameters, and the delay estimation accuracy is demonstrated by evaluating multiple circuits. Experiments results indicate that the IGD based approach provides a high matching against HSPICE Monte Carlo simulation results, with an average error less than 1.9% and 1.2% for the 8-bit Ripple Carry Adder(RCA), and 8-bit De-Multiplexer(DEMUX) and Multiplexer(MUX) respectively