※ 引述《moonshade (翻滚吧！猫咪)》之铭言： : 既然都这样说了 : 有空的看看吧... : This paper presents a greedy refinement technique for IR-drop : reduction. The authors applied a fast IR drop estimation method : of a existing work to locate the row with maximum IR-drop, and (from?) an : reduces it's IR-drop by moving the cell in that row to other (to) reduce its cell(s) : row's with less IR-drop. rows : 1. : a). : The resistive model for standard cells in this paper is equivalent to : a VCCS source with : I = Ps/V + k*V (derived from Equation (4) and (5) : considering that P=V*I), where Ps is standard power consumption, : I is the value of the current source, : V is the voltage difference from power pin and ground pin and : k is a parameter related to clock rate and etc.. In some situations, : for example, if the static power consumption larger than dynamic power is : consumption, the current will increase while V decrease, which contradicts decreases ^^^^^for what? : with the behavior of any existed standard cell. : b). The resistive model SHOULD not be directed grounded. For constant : current model the voltage cross a cell and the output current is , : independent to each other. : The current source can be grounded without any problem. But it is not : the case in the resistive or other any VCCS model. Those resistors : should connected to both the ground network and the power network. connect : Due to the above reasons, the reviewer doubts the accuracy of this model. : The experimental results should have be verified with commercial tools, : such as powerrail, voltagestorm, or HSPICE. : 2. The contribution of this paper is Theorem 1. But it is trivial for : electronic circuits. Removing a standard cell which consumes current : will no doubt reduce IR-drop. If removing a standard cell out of a : row will increase IR-drop, the model of the power/ground network is : definitely wrong. --

※ 发信站: 批踢踢实业坊(ptt.cc) ◆ From: 140.112.5.65