home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.lsi
- Path: sparky!uunet!psgrain!percy!klic!keithl
- From: keithl@klic.rain.com (Keith Lofstrom)
- Subject: Re: Die aspect ratio.
- Message-ID: <1992Aug16.085641.29799@klic.rain.com>
- Organization: Keith Lofstrom Integrated Circuits
- References: <16erivINNfir@agate.berkeley.edu>
- Date: Sun, 16 Aug 1992 08:56:41 GMT
- Lines: 48
-
- In article <16erivINNfir@agate.berkeley.edu> krste@ICSI.Berkeley.EDU (Krste Asanovic) writes:
- >
- >It seems unlikely that an irregular design such as a processor or
- >cache controller would map naturally to a square die floorplan. Is
- >there some inherent yield or manufacturability advantage to having a
- >square die? I can think of reticle size and maybe efficient wafer
- >tiling as two reasons.
- >
- >Rectangular die give more perimeter per unit area if a design is pad
- >limited. What's the largest sensible/known aspect ratio? (We've done
- >one processor die with an aspect ratio of 2.3:1).
-
- If a die is core limited, a square die will be the most efficient. Wafer
- tiling is more efficient, as well. The photo process is better towards
- the center of the reticle, and a square die will yield and align better
- than a rectangular one, due to accumulated errors along the long axis
- of the die.
-
- In some cases, due to step field size, cad tools, probe card limits, etc,
- it is difficult to make one dimension larger than a certain size. This
- means large die tend to be square. Also, some aspects of driver design
- are easier if you have to do calculations of wire length, etc, for one
- side dimension rather than two.
-
- Most standard open-tool packages are designed around square die, and with
- minimum wirebond lengths of 20 mils and maximum wirebond lengths of 120 mils,
- a die with more than 200 mils difference in side lengths will need a custom
- package. Even within the 200 mil limit, the long bond wires will be more
- inductive on one side than the other, leading to more design work.
-
- My last two projects were perfectly square, into a square package. I
- worry about assembly mistakes (accidental rotation of the die) though
- worst case we just grind a new pin 1 mark on the plastic package, or spec
- 4 different varieties ;-).
-
- The largest aspect ratio I have heard about were some Reticon CCD linear
- imaging arrays - these were quite long and skinny. Tektronix made some CCD
- waveform sampling arrays that were pretty long, but those had to be wide
- enough so they wouldn't fall apart during scribe and break. I ran an Explorer
- Scout group that used some of the "left-over" width for some student test
- projects - high school kids designing tiny (200 gate) digital circuits,
- kinda fun.
-
- Keith
- --
- Keith Lofstrom keithl@klic.rain.com Voice (503)-520-1993
- KLIC --- Keith Lofstrom Integrated Circuits --- "Your Ideas in Silicon"
- Design Contracting in Bipolar and CMOS - Analog, Digital, and Power ICs
-