CPU屏蔽                        
                
                                
                        
                            中央处理器                        
                
                                
                        
                            CPU核心电压                        
                
                                
                        
                            计算机科学                        
                
                                
                        
                            节点(物理)                        
                
                                
                        
                            嵌入式系统                        
                
                                
                        
                            电气工程                        
                
                                
                        
                            电子工程                        
                
                                
                        
                            材料科学                        
                
                                
                        
                            工程类                        
                
                                
                        
                            计算机硬件                        
                
                                
                        
                            电压                        
                
                                
                        
                            交流电源                        
                
                                
                        
                            结构工程                        
                
                                
                        
                            电压优化                        
                
                        
                    
            作者
            
                Rongmei Chen,Melina Lofrano,Gioele Mirabelli,Giuliano Sisto,Sheng Yang,Anne Jourdain,Filip Schleicher,A. Veloso,Odysseas Zografos,Pieter Weckx,Gaspard Hiblot,Geert Van der Plas,Geert Hellings,Julien Ryckaert,Eric Beyne            
         
            
    
            
            标识
            
                                    DOI:10.1109/iedm45625.2022.10019349
                                    
                                
                                 
         
        
                
            摘要
            
            In this work, we comprehensively evaluate the impact of using backside power delivery network (BSPDN) for CPU of 2D and 3D designs at A14 node from the aspects of power, performance, area and thermal (PPAT) compared to identical designs of conventional front-side (FS) PDN. 2D BSPDN power consumption is 57% smaller than that of 2D FSPDN at iso-CPU frequency. 3D CPU-on-CPU PDN power consumption shows ~3.8 totally (or ~1.9/CPU) times the 2D FSPDN counterpart. Ring oscillator evaluation shows that logic gates in the IR-drop hotspot region of CPU has worst performance loss <9% and >16% for BSPDN and FSPDN based CPUs respectively while the 3D top CPU can suffer from 25% performance loss. BSPDN based CPU area can be scaled down by 8% compared with the FSPDN counterpart with similar power and performance after physical design and PPA evaluation. Due to the thinning of substrate thickness and the BSPDN process, BSPDN based CPU local temperature can be ~40% more than the FSPDN counterpart, illustrating the importance of considering the heat dissipation for the chips with BSPDN. For the CPU-on-CPU 3D IC with power source/package on the bottom and cooler on the top, the top CPU die has 39% more IR drop than the bottom one while the later has 17% more temperature than the former. This opposite trend may induce additional challenge of power integrity and thermal reliability co-design and optimization for 3D ICs with BSPDN.
         
            
 
                 
                
                    
                    科研通智能强力驱动
Strongly Powered by AbleSci AI