Sarmad, once again I think you have not grasped typical time frames for processor production. When a new product, design or process, is being developed, it first must have the logic designed and verified and a layout for production completed. That is referred to as "tapeout" (because it used to involve recording the data to a tape). From that masks are made and first silicon is produced using rocket lots. Sometimes that actually functions, sometimes not. The individual layers are scrutinized and new masks are cut to fix whatever problems are found. After a few iterations the product appears to work in the lab, usually at far less than desirable clockspeeds and unmanufacturable yields. Nonetheless these chips can be sent out as engineering samples and the product moved to manufacture. That marks the start of the production ramp. The production lab plays with various parameters, timings and chemical applications to fine tune the process to produce better yields and higher frequency chips. Since the early wafers produce little or no usable output, most companies limit their number to extract the maximum information at the minimum cost. That means that the first couple months of production don't amount to much but learning (Intel is an exception as they are more aggressive with early wafer starts than most anyone else).
Typically after a few months yields on the new process are approaching the level where usable die per wafer is near parity with the old process (still a lower percentage because the shrink yields many more candidates per wafer). At that point there is no loss in output from converting wafer starts from the old process to the new (but still a risk that the as yet unreleased chips will have some defect that will only appear in use). About that time you want a limited launch of the new chip so that any problems will be somewhat confined and discovered before you are totally screwed.
Now add in about 12 weeks from wafer start to finished processor and you have a handle on the typical time scale for a ramp.
In April AMD was producing and sampling 65nm chips. In June they began the production ramp, and 6 months later they have enough product for a launch and claim that per wafer good die yields are higher on the new process than the old.
So they have pretty much been on schedule since April. Should they have been ahead of that schedule then? Initially it was expected that Fab 36 would ramp first at 65nm instead of 90, so that was a 6 month setback. In February and even March, AMD was above 40 so I guess it is fair to say there were some disappointments between then and the July bottom at 16 and change. Some were process related, some related to competitive pressure as Intel dealt with its inventory mess, some related to delays in Vista, and some to the strength of CMW. I actually sold some shares at 40 but not nearly enough, and avoided posting a general opinion most of the year because I kind of figured it had made a top, though I lacked the confidence to unload my whole position. I didn't figure they could get back to 17, and still think that was overdone and reflected the unexpected (and still questionable to me) ATI acquisition. Nonetheless I bought more at the bottom and think AMD is good value at this level.
e |
