The effect of manufacturing process variations has become a major issue regarding the estimation of circuit delay and power dissipation, and will gain more importance in the future as device scaling continues in order to satisfy market place demands for circuits with greater performance and functionality per unit area. Statistical modelling and analysis approaches have been widely used to reflect the effects of a variety of variational process parameters on system performance factor which will be described as probability density functions (PDFs). At present most of the investigations into statistical models has been limited to small circuits such as a logic gate. However, the massive size of present day electronic systems precludes the use of design techniques which consider a system to comprise these basic gates, as this level of design is very inefficient and error prone. This thesis proposes a methodology to bring the effects of process variation from transistor level up to architectural level in terms of circuit delay and leakage power dissipation. Using a first order canonical model and statistical analysis approach, a statistical cell library has been built which comprises not only the basic gate cell models, but also more complex functional blocks such as registers, FIFOs, counters, ALUs etc. Furthermore, other sensitive factors to the overall system performance, such as input signal slope, output load capacitance, different signal switching cases and transition types are also taken into account for each cell in the library, which makes it adaptive to an incremental circuit design. The proposed methodology enables an efficient analysis of process variation effects on system performance with significantly reduced computation time compared to the Monte Carlo simulation approach. As a demonstration vehicle for this technique, the delay and leakage power distributions of a 2-stage asynchronous micropipeline circuit has been simulated using this cell library. The experimental results show that the proposed method can predict the delay and leakage power distribution with less than 5% error and at least 50,000 times faster computation time compare to 5000-sample SPICE based Monte Carlo simulation. The methodology presented here for modelling process variability plays a significant role in Design for Manufacturability (DFM) by quantifying the direct impact of process variations on system performance. The advantages of being able to undertake this analysis at a high level of abstraction and thus early in the design cycle are two fold. First, if the predicted effects of process variation render the circuit performance to be outwith specification, design modifications can be readily incorporated to rectify the situation. Second, knowing what the acceptable limits of process variation are to maintain design performance within its specification, informed choices can be made regarding the implementation technology and manufacturer selected to fabricate the design.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:588264 |
| Date | January 2013 |
| Creators | Ni, Chenxi |
| Publisher | University of Newcastle upon Tyne |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://hdl.handle.net/10443/1938 |
Page generated in 0.0017 seconds