📄 ch03.6.htm
字号:
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML EXPERIMENTAL 970324//EN">
<HTML>
<HEAD>
<META NAME="GENERATOR" CONTENT="Adobe FrameMaker 5.5/HTML Export Filter">
<TITLE> 3.6 Gate-Array Design</TITLE></HEAD><!--#include file="top.html"--><!--#include file="header.html"-->
<DIV>
<P>[ <A HREF="CH03.htm">Chapter start</A> ] [ <A HREF="CH03.5.htm">Previous page</A> ] [ <A HREF="CH03.7.htm">Next page</A> ]</P><!--#include file="AmazonAsic.html"--><HR></DIV>
<H1 CLASS="Heading1">
<A NAME="pgfId=94550">
</A>
3.6 <A NAME="14984">
</A>
Gate-Array Design</H1>
<P CLASS="BodyAfterHead">
<A NAME="pgfId=73082">
</A>
Each logic cell or macro in a gate-array library is predesigned using fixed tiles of transistors known as the <A NAME="marker=73083">
</A>
<SPAN CLASS="Definition">
gate-array base cell</SPAN>
(or just <A NAME="marker=73084">
</A>
<SPAN CLASS="Definition">
base cell</SPAN>
). We call the arrangement of base cells across a whole chip in a complete gate array the <A NAME="marker=73085">
</A>
<SPAN CLASS="Definition">
gate-array base</SPAN>
(or just <SPAN CLASS="Definition">
base</SPAN>
<A NAME="marker=73086">
</A>
). ASIC vendors offer a selection of bases, with a different total numbers of transistors on each base. For example, if our ASIC design uses 48k equivalent gates and the ASIC vendor offers gate arrays bases with 50k-, 75k-, and 100k-gates, we will probably have to use the 75k-gate base (because it is unlikely that we can use 48/50 or 96<SPAN CLASS="Symbol">
</SPAN>
percent of the transistors on the 50k-gate base).</P>
<P CLASS="Body">
<A NAME="pgfId=99187">
</A>
We isolate the transistors on a gate array from one another either with thick field oxide (in the case of oxide-isolated gate arrays) or by using other transistors that are wired permanently off (in gate-isolated gate arrays). Channeled and channelless gate arrays may use either gate isolation or oxide isolation.</P>
<P CLASS="Body">
<A NAME="pgfId=73092">
</A>
Figure <A HREF="#11168" CLASS="XRef">
3.14</A>
(a) shows a base cell for a <A NAME="marker=73091">
</A>
<SPAN CLASS="Definition">
gate-isolated gate array</SPAN>
. This base cell has two transistors: one <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-channel and one <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-channel. When these base cells are placed next to each other, the <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-diffusion and <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-diffusion layers form continuous strips that run across the entire chip broken only at the poly gates that cross at regularly spaced intervals (Figure 3.14b). The metal interconnect spacing determines the separation of the transistors. The metal spacing is determined by the design rules for the metal and contacts. In Figure <A HREF="#11168" CLASS="XRef">
3.14</A>
(c) we have shown all possible locations for a contact in the base cell. There is room for 21 contacts in this cell and thus room for 21 interconnect lines running in a horizontal direction (we use m1 running horizontally). We say that there are 21 <SPAN CLASS="Definition">
horizontal tracks</SPAN>
<A NAME="marker=99170">
</A>
in this cell or that the cell is 21 tracks high. In a similar fashion the space that we need for a vertical interconnect (m2) is called a <A NAME="marker=99171">
</A>
<SPAN CLASS="Definition">
vertical track</SPAN>
. The horizontal and vertical track widths are not necessarily equal, because the design rules for m1 and m2 are not always equal.</P>
<P CLASS="Body">
<A NAME="pgfId=225106">
</A>
We isolate logic cells from each other in gate-isolated gate arrays by connecting transistor gates to the supply bus—hence the name, <A NAME="marker=225107">
</A>
<SPAN CLASS="Definition">
gate isolation</SPAN>
. If we connect the gate of an <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-channel transistor to <SPAN CLASS="EquationVariables">
V</SPAN>
<SUB CLASS="SubscriptVariable">
SS </SUB>
, we isolate the regions of <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-diffusion on each side of that transistor (we call this an <SPAN CLASS="Definition">
isolator transistor</SPAN>
<A NAME="marker=225108">
</A>
or device, or just isolator). Similarly if we connect the gate of a <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-channel transistor to <SPAN CLASS="EquationVariables">
V</SPAN>
<SUB CLASS="SubscriptVariable">
DD</SUB>
, we isolate adjacent <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-diffusion regions. </P>
<TABLE>
<TR>
<TD ROWSPAN="1" COLSPAN="1">
<P CLASS="TableFigure">
<A NAME="pgfId=225174">
</A>
<IMG SRC="CH03-28.gif" ALIGN="BASELINE">
</P>
</TD>
</TR>
<TR>
<TD ROWSPAN="1" COLSPAN="1">
<P CLASS="TableFigureTitle">
<A NAME="pgfId=225178">
</A>
FIGURE 3.14 <A NAME="11168">
</A>
The construction of a gate-isolated gate array. (a) The one-track-wide base cell containing one <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-channel and one <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-channel transistor. (b) Three base cells: the center base cell is being used to isolate the base cells on either side from each other. (c) A base cell including all possible contact positions (there is room for 21 contacts in the vertical direction, showing the base cell has a height of 21 tracks).</P>
</TD>
</TR>
</TABLE>
<P CLASS="Body">
<A NAME="pgfId=225182">
</A>
Oxide-isolated gate arrays often contain four transistors in the base cell: the two <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-channel transistors share an <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-diffusion strip and the two <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-channel transistors share a <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-diffusion strip. This means that the two <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-channel transistors in each base cell are electrically connected in series, as are the <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-channel transistors. The base cells are isolated from each other using <A NAME="marker=225183">
</A>
<SPAN CLASS="Definition">
oxide isolation</SPAN>
. During the fabrication process a layer of the thick field oxide is left in place between each base cell and this separates the <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-diffusion and <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-diffusion regions of adjacent base cells.</P>
<P CLASS="Body">
<A NAME="pgfId=99198">
</A>
Figure <A HREF="#22766" CLASS="XRef">
3.15</A>
shows an <A NAME="marker=73257">
</A>
<SPAN CLASS="Definition">
oxide-isolated gate array</SPAN>
. This cell contains eight transistors (which occupy six vertical tracks) plus one-half of a single track that contains the well contacts and substrate connections that we can consider to be shared by each base cell.</P>
<TABLE>
<TR>
<TD ROWSPAN="1" COLSPAN="1">
<P CLASS="TableFigure">
<A NAME="pgfId=73291">
</A>
</P>
<DIV>
<IMG SRC="CH03-29.gif">
</DIV>
</TD>
</TR>
<TR>
<TD ROWSPAN="1" COLSPAN="1">
<P CLASS="TableFigureTitle">
<A NAME="pgfId=73294">
</A>
FIGURE 3.15 <A NAME="22766">
</A>
An oxide-isolated gate-array base cell. The figure shows two base cells, each containing eight transistors and two well contacts. The <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-channel and <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-channel transistors are each 4 tracks high (corresponding to the width of the transistor). The leftmost vertical track of the left base cell includes all 12 possible contact positions (the height of the cell is 12 tracks). As outlined here, the base cell is 7 tracks wide (we could also consider the base cell to be half this width).</P>
</TD>
</TR>
</TABLE>
<P CLASS="Body">
<A NAME="pgfId=77033">
</A>
Figure <A HREF="#39811" CLASS="XRef">
3.16</A>
shows a base cell in which the gates of the <SPAN CLASS="Emphasis">
n</SPAN>
-channel and <SPAN CLASS="Emphasis">
p</SPAN>
-channel transistors are connected on the polysilicon layer. Connecting the gates in poly saves contacts and a metal interconnect in the center of the cell where interconnect is most congested. The drawback of the preconnected gates is a loss in flexibility in cell design. Implementing memory and logic based on transmission gates will be less efficient using this type of base cell, for example. </P>
<TABLE>
<TR>
<TD ROWSPAN="1" COLSPAN="1">
<P CLASS="TableFigure">
<A NAME="pgfId=76962">
</A>
<IMG SRC="CH03-30.gif" ALIGN="BASELINE">
</P>
</TD>
</TR>
<TR>
<TD ROWSPAN="1" COLSPAN="1">
<P CLASS="TableFigureTitle">
<A NAME="pgfId=76965">
</A>
FIGURE 3.16 <A NAME="39811">
</A>
This oxide-isolated gate-array base cell is 14 tracks high and 4 tracks wide. VDD (tracks 3 and 4) and GND (tracks 11 and 12) are each 2 tracks wide. The metal lines to the left of the cell indicate the 10 horizontal routing tracks (tracks 1, 2, 5–10, 13, 14). Notice that the <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-channel and <SPAN CLASS="Emphasis">
n</SPAN>
-channel polysilicon gates are tied together in the center of the cell. The well contacts are short, leaving room for a poly cross-under in each base cell.</P>
</TD>
</TR>
</TABLE>
<P CLASS="Body">
<A NAME="pgfId=77100">
</A>
Figure <A HREF="#27453" CLASS="XRef">
3.17</A>
shows the metal <SPAN CLASS="Definition">
personalization</SPAN>
<A NAME="marker=99281">
</A>
for a D flip-flop macro in a gate-isolated gate array using a base cell similar to that shown in Figure <A HREF="#11168" CLASS="XRef">
3.14</A>
(a). This macro uses 20 base cells, for a total of 40 transistors, equivalent to 10 gates. </P>
<TABLE>
<TR>
<TD ROWSPAN="1" COLSPAN="1">
<P CLASS="TableFigure">
<A NAME="pgfId=77096">
</A>
<IMG SRC="CH03-31.gif" ALIGN="BASELINE">
</P>
</TD>
</TR>
<TR>
<TD ROWSPAN="1" COLSPAN="1">
<P CLASS="TableFigureTitle">
<A NAME="pgfId=77099">
</A>
FIGURE 3.17 <A NAME="27453">
</A>
An example of a flip-flop macro in a gate-isolated gate-array library. Only the first-level metallization and contact pattern (the personalization) is shown on the right, but this is enough information to derive the schematic. The base cell is shown on the left. This macro is 20 tracks wide.</P>
</TD>
</TR>
</TABLE>
<P CLASS="Body">
<A NAME="pgfId=87719">
</A>
The gates of the base cells shown in Figures <A HREF="#11168" CLASS="XRef">
3.14</A>
–<A HREF="#39811" CLASS="XRef">
3.16</A>
are bent. The <A NAME="marker=87720">
</A>
<SPAN CLASS="Definition">
bent gate</SPAN>
allows contacts to the gates to be placed on the same grid as the contacts to diffusion. The polysilicon gates run in the space between adjacent metal interconnect lines. This saves space and also simplifies the routing software. </P>
<P CLASS="Body">
<A NAME="pgfId=39216">
</A>
There are many trade-offs that determine the gate-array base cell height. One factor is the number of wires that can be run horizontally through the base cell. This will determine the capacity of the routing channel formed from an unused row of base cells. The base cell height also determines how easy it is to wire the logic macros since it determines how much space for wiring is available inside the macros.</P>
<P CLASS="Body">
<A NAME="pgfId=92632">
</A>
There are other factors that determine the width of the base-cell transistors. The widths of the <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-channel and <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-channel transistors are slightly different in Figure <A HREF="#11168" CLASS="XRef">
3.14</A>
(a). The <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-channel transistors are 6 tracks wide and the <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-channel transistors are 5 tracks wide. The ratio for this gate-array library is thus approximately 1.2. Most gate-array libraries are approaching a ratio of 1.</P>
<P CLASS="Body">
<A NAME="pgfId=159473">
</A>
ASIC designers are using ever-increasing amounts of <A NAME="marker=159472">
</A>
RAM on gate arrays. It is inefficient to use the normal base cell for a static RAM cell and the size of RAM on an embedded gate array is fixed. As an alternative we can change the design of the base cell. A base cell designed for use as RAM has extra transistors (either four—two <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-channel and two <SPAN CLASS="EmphasisPrefix">
p</SPAN>
-channel—or two <SPAN CLASS="EmphasisPrefix">
n</SPAN>
-channel; usually minimum width) allowing a six-transistor RAM cell to be built using one base cell instead of the two or three that we would normally need. This is one of the advantages of the <SPAN CLASS="Definition">
CBA</SPAN>
<A NAME="marker=193240">
</A>
<A NAME="marker=193241">
</A>
(cell-based array) base cell shown in Figure <A HREF="#36763" CLASS="XRef">
3.18</A>
. </P>
<TABLE>
<TR>
<TD ROWSPAN="1" COLSPAN="1">
<P CLASS="TableFigure">
<A NAME="pgfId=187411">
</A>
<IMG SRC="CH03-32.gif" ALIGN="BASELINE">
</P>
</TD>
</TR>
<TR>
<TD ROWSPAN="1" COLSPAN="1">
<P CLASS="TableFigureTitle">
<A NAME="pgfId=187414">
</A>
FIGURE 3.18 <A NAME="36763">
</A>
The SiARC/Synopsys cell-based array (CBA) basic cell.</P>
</TD>
</TR>
</TABLE>
<HR><P>[ <A HREF="CH03.htm">Chapter start</A> ] [ <A HREF="CH03.5.htm">Previous page</A> ] [ <A HREF="CH03.7.htm">Next page</A> ]</P></BODY>
<!--#include file="Copyright.html"--><!--#include file="footer.html"-->
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -