Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : LSI Corporation -- Ignore unavailable to you. Want to Upgrade?

To: Grand Poobah who wrote (12462)	5/16/1998 6:35:00 AM
From: Grand Poobah	Read Replies (2) \| Respond to of 25814

Shane, Talking about yields and our previous mentions of the importance of EDA tools and the idea that it's not that easy to make a SOC got me to thinking it might be helpful to have a concise description of what's involved in the design and production of an IC. This has nothing to do directly with the stock price of LSI, but I thought you and possibly others might be interested in how the product is made, if you don't already know. And I hope it helps to shed a little light on the complexity of the process and what all besides the demand for the product really has to go right for a design to work. Those who are more intimate with the design process or otherwise have more experience, please feel free to correct my errors and omissions. The design, of course, starts out as a market-driven idea; a customer, or Wilf, or someone else says wouldn't it be great to have a chip that would do X, and by Jove, I think LSI just might be able to make one. They throw together a team of design engineers into a dark hole with some workstations and a few snacks, and boom, six months or a year later, out pops a design. Actually I never was a design engineer, but I do know that they use their creativity and training and EDA tools to turn the idea into the gates, latches, and registers that make up the "Logic" part of LSI Logic. Then the place-and-route tool (another EDA tool) turns these digital and analog logic concepts into a collection of interconnected transistors at specific locations. This produces a layout and netlist which simulate the behavior of the chip and are checked for functionality (using an EDA tool). They then have to be verified (by an EDA tool) against a set of design rules specific to the kind of design and especially the fabrication process. The layout is a collection of layers of polygons which describe the chip physically. The netlist contains the information which describes the chip electrically. The design rules specify things like the size of the transistors and the width and spacing of the interconnecting wires. The netlist and layout are computer files which are many megabytes or sometimes even a gigabyte or two in size. Once the design is verified, the layout file is sent to the photomask company for generation of the masks. There can be typically up to 16 to 20 masks per design. These masks are then sent to the fab and each one is used to generate a layer or layers of the design on the silicon wafers, using photolithography. The pattern from the mask is repeated on the wafer, using step-and-repeat equipment (steppers) and exposing nine (or sometimes four) chips at a time, until the whole wafer is done. A wafer typically holds a few hundred chips. The chips on the wafer are then "probed", or tested, to determine whether they are good or bad. This is done using a test pattern which was generated (using an EDA tool) during the design process. There are several different basic methods of testing, but in general they give the chip certain sets of inputs and check to see if the expected outputs are generated by the chip. This test typically takes a few seconds per chip and may take as long as a minute. The bad chips are marked with a dot of ink. The wafers then go to the packaging company, where the wafer is sawed, or "diced", into the individual chips, and the good chips from wafer test are put into plastic or ceramic packages which protect them from the environment and facilitate connection to the printed circuit board they will placed on in the eventual application. These packaged chips are then tested for a final time ("final test"), put in a tray and a box, and shipped to the customer for that revenue we all like to see (once all the debug of a new design has been done). G.P.

To: Grand Poobah who wrote (12462)	5/18/1998 6:30:00 PM
From: Tony Viola	Read Replies (1) \| Respond to of 25814

PB, Re: "I'm curious about where you got your info on relative yield numbers," and Re: I'm curious about where you got your info on relative yield numbers:" Industry. Well, from industry people at conferences I have attended, like IEEE Computer Elements Workshops. IBM specifically is very open about their processes, and even about their yields. And why not, they really are so good that they've got nothing to hide. I haven't seen anything in print from them, like I said, because companies are loathe to put yield numbers down on paper for the public to see. But, they will talk about them openly. Another thing is that IBM and others have long been convinced that poor yield leads only to poor reliability. I went around and around on this with some Intel haters over on the Intel thread. Finally, someone asked Paul Engel, the best by far expert over there, if I was all wet or not. He said no, I was exactly right. One more thing about IBM..they have what they call maverick wafers. What they do is that, as they are stepping across the rows and columns on wafers, the computer controlled tester is keeping track of yield. If the tester gets so far into the wafer (don't know what percentage of it) and the yield is running lower than a set standard, they stop testing that one and throw it away! Now, we've heard that certain companies have had yield problems...severe ones, but have yielded the "good" dice, packaged and shipped them. Don't buy a computer built with these companies' chips. Don't ask me who, because my Mom told me if I can't say something good about someone, don't say anything at all. Tony