From WikiChip
Difference between revisions of "mtr-mm²"

(Overview)
(Overview)
Line 3: Line 3:
  
 
== Overview ==
 
== Overview ==
The metric was proposed by Mark Bohr, [[Intel]]'s director of process architecture and integration, in 2017. Bohr stated the metric is actually a resurrection of a similar metric that had been abandoned in the past. The metric attempts to take into account second order design rules that the simpler density metrics such as [[CPP x MMP]] and [[CPP x MMP x Tracks]] cannot capture because they do not represent a typical [[standard cell]]. It should be noted that the metric explicitly [[SRAM]] cell sizes because of the large variance in SRAM-to-logic ratio between popular chips. Therefore, SRAM cell sizes should be reported separately alongside the MTr/mm² unit.
+
The metric was proposed by Mark Bohr, [[Intel]]'s director of process architecture and integration, in 2017. Bohr stated the metric is actually a resurrection of a similar metric that had been abandoned in the past. The metric attempts to take into account second order design rules that the simpler density metrics such as [[CPP x MMP]] and [[CPP x MMP x Tracks]] cannot capture because they do not represent a typical [[standard cell]]. It should be noted that the metric explicitly excludes [[SRAM]] cell sizes because of the large variance in SRAM-to-logic ratio between popular chips. Furthermore, the metal pitch does not play a role in limiting SRAM as it does with other standard cells. Therefore, SRAM cell sizes should be reported separately alongside the MTr/mm² unit.
  
 
Two [[standard cell|cells]] that are found in every [[standard library]] are used: a very small 2-input [[NAND]] cell consisting of just four [[transistors]] and a very large [[scan flip-flop]] cell. The NAND2 is weighted as 0.6 while the SFF is weighted as 0.4.
 
Two [[standard cell|cells]] that are found in every [[standard library]] are used: a very small 2-input [[NAND]] cell consisting of just four [[transistors]] and a very large [[scan flip-flop]] cell. The NAND2 is weighted as 0.6 while the SFF is weighted as 0.4.

Revision as of 10:16, 18 November 2017

MTr/mm² (mega-transistor per squared millimeter) is a transistor density unit that attempts to quantify a process node. The metric makes use of a weighted system consisting of two typical standard cells found in most libraries - a very small NAND2 cell and a very large SFF cell.

Overview

The metric was proposed by Mark Bohr, Intel's director of process architecture and integration, in 2017. Bohr stated the metric is actually a resurrection of a similar metric that had been abandoned in the past. The metric attempts to take into account second order design rules that the simpler density metrics such as CPP x MMP and CPP x MMP x Tracks cannot capture because they do not represent a typical standard cell. It should be noted that the metric explicitly excludes SRAM cell sizes because of the large variance in SRAM-to-logic ratio between popular chips. Furthermore, the metal pitch does not play a role in limiting SRAM as it does with other standard cells. Therefore, SRAM cell sizes should be reported separately alongside the MTr/mm² unit.

Two cells that are found in every standard library are used: a very small 2-input NAND cell consisting of just four transistors and a very large scan flip-flop cell. The NAND2 is weighted as 0.6 while the SFF is weighted as 0.4.


Equation Transistor Count slash mm square equals 0.6 times StartFraction NAND 2 trace Count Over NAND 2 Cell Area EndFraction plus 0.4 times StartFraction Scan Flip Flop trace Count Over Scan Flip Flop Cell Area EndFraction


mtr-mm2.svg

Example

Intel's 10 nm process consists of a minimum metal pitch of 36nm with 7.56 tracks for a cell height of 272 nanometers. Additionally, the process has a poly pitch of 54 nm. For 0.6 NAND2 + 0.4 SFF, Intel's 10nm has a density of 102.9 MTr/mm² along with a high-density 6T SRAM measuring 0.0312 µm². Note that Intel itself reported their 10nm at 100.8 MTr/mm².