| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
I’d like address the question about vias in regards to the ODM design model. Backplanes may receive some TLC. It is often a different story when I comes to the design and manufacture of compute blades that are produced by ODMs. The ODM is the model for the mid/low end system. Mid/low end systems and the like were suggested in mellitz_01_0112, frazier_01_0112.pdf, Nowell_01_0112.pdf, goergen_01_0112.pdf, and even on page 4 of mohan_01_0112.pdf. I really don’t think this is an imagined market. Now for ODMs. In my experience, getting ODMs to do simulation is like asking: do you like root canals? For ODMs, vias are really just a part in a layout data base. If it is not in the data base it does not exist. Kent Lusted told me in a sidebar that even if there may be a good via “part” in the data base, in all likelihood it can’t be shared to other products because of IP. We see the ODM model becoming the volume model for mid/low end servers. The model is enables the snowballing of technology. I my opinion not enabling this market would significantly reduce the market potential. Like I said, from a computer/server system manufacturing perspective, I don’t care what PHY you use as long as you support this market. Right now PAM-4 is trending this way. I’m open for other suggestions here. There was another comment about crosstalk and vias being bad for PAM-4. The real problem is what do I do about 40-60 dB of loss at 12.9Ghz suggested yesterday in many of the presentations. I don’t think repeaters are the answer. Richard Mellitz Intel Corporation |