Research And Verification On STA Method Of GPU On 16nm Process

Static Timing Analysis(STA)is a quite significant part of large scale integrated circuit design which verify the timing accuracy of the design and determine whether the circuit could work at the required clock frequency.Due to the on-chip variation in the chip manufacturing process,the device patterns and working condition are dissimilar with simulation,it has big influnence on STA.So it is necessary for back-end design to research on accurately and efficiently analysis method based on different chip.Based on the timing anlysis requirement of the Graphics Processing Unit(GPU)chip on 16 nm process,this paper contained the study concerned the impact of the on-chip variation on static timing analysis method,then explored how to apply different method in GPU and compared the result.Finally found the more accurate and efficient timing analysis methods in the back-end physical design of GPU to improve the speed of timing closure.In this paper,the research based on EDA tool Innovus and Prime Time.First of all,we used the on chip variation analysis method to explore the on-chip variation of the 16 nm GPU chip compared with the research in traditional best corner-worst corner analysis method.Afterwards,in order to solve the issues about the slow speed of timing closure with on chip variation method,we used advanced on chip variation analysis method.Finally,we studied the statistical on chip variation method which is a new method in 16 nm process so as to accelerate the accuracy and speed of timing convergence.Simultaneously,this paper explored and applied the best static timing analysis process based on in back-end flow of GPU with 16 nm node.In the end,compared the timing results of the three STA analysis methods indicated that on chip variation method was capable of calculating the on-chip bias during static timing analysis,but it made the total negative slack 61.4% worse than best corner-worst corner method,and the runtime of timing optimization increasing nearly 20%,so the speed of timing closure became slower.Advanced on chip variation method is more accurate than on chip variation,and the total negative slack was optimized by 30.4%,which revealed that advanced on chip variation is more benefit to timing convergence than on chip variation.However,advanced on chip variation method takes about 0.87 time longer optimization runtime which would cost more extra time to design the chip.Compared the worst negative slack,total negative slack as well as the runtime of timing optimization with advanced on chip variation,found that the worst negative slack of statistical on chip variation is about 29.1% better than advanced on chip variation,and total negative slack is about 46.7% better than advanced on chip variation,total runtime of timing optimization reduced about 54.9%.It revealed that the statistical on chip variation method is more suitable static timing analysis method of GPU on 16 nm process and more accurate than advanced on chip variation analysis method based on the abnormal timing analysis results and correlation result with Prime Time.Therefore,it verified that the this parper achieved best timing analysis of GPU on 16 nm process.