Page 362 - A Practical Guide from Design Planning to Manufacturing
P. 362

332   Chapter Eleven

        the first time. In the best scenario, the chips may behave exactly as sim-
        ulations predicted they would. In the worst case, the only result of
        applying power may be smoke from a suddenly melting circuit board.
        Most first silicon designs fall somewhere between these extremes, being
        somewhat functional but not behaving exactly as expected. Silicon debug
        and test is the story of what happens next.
          Even design teams lucky enough to have their first chips perform
        perfectly are left with the nagging question, “Are we ready to ship to cus-
        tomers?” If the product successfully runs 10 programs, does that prove
        the design is correct? Is that sufficient to begin sales? If not, is testing
        a hundred programs sufficient, or a thousand? This is the job of post-
        silicon validation, testing of finished parts to find design bugs. A silicon
        design bug is any unexpected behavior from the part, and first silicon
        chips inevitably have some bugs.
          Pre-silicon validation attempts to find and correct as many bugs as
        possible before first silicon is created, but the slow speed of pre-silicon
        simulations requires years to test only minutes of actual operation. In
        hardware, millions of transistors are switching in parallel to determine
        the behavior of the processor. In simulation, the behavior of one imag-
        inary transistor must first be determined before simulating the next and
        then the next. This serialization of transistor behavior dooms simula-
        tions to being millions of times slower than hardware operation. Because
        of this difference in speed, the shortest time to market is usually not to
        try and produce a perfect first silicon design. Instead, the first silicon
        chips will be produced at the point when it is believed that post-silicon
        validation can find the remaining bugs and bring the product to market
        more quickly than continued pre-silicon simulation.
          When bugs are detected, their cause must be found and fixed. The
        process of finding the root cause and eliminating design flaws in silicon is
        called silicon debug. The time spent in silicon debug has a direct impact
        on how quickly a product ships to customers; for a new processor design,
        debug may last from a few months to a couple of years. 1,2  For products that
        have life cycles of a few years at best, reaching the market a few months
        early because of effective silicon debug has a huge impact on profits.
          Even after design flaws have been fixed, there will always be some die
        that do not function properly because of manufacturing defects. If all
        manufacturing defects could be eliminated, the process dimensions would
        be reduced to cut costs and increase performance until some defects
        began to appear. As a result, there will always be some nonfunctional die,
        and there must be a method for quickly and reliably identifying these die.



          1                     ®
          Carbine and Derek, “Pentium Pro Processor Design.”
          2                     ®       ®
          Bentley, “Validating the Intel Pentium 4.”
   357   358   359   360   361   362   363   364   365   366   367