Well, the feedback from the gentle readers
2112's post sounded awfully cryptic
and not-so-gentle readers
Quoted, because it seems like you actually believe that detailed information about TSMC's processes exists in the modded cgminer.
is that my "Teach yourself IC design,fabrication&test in 21 minutes" lecture is too hard and sounds like black magic.
I'm going to quote single linear thought from my post to maybe make it easier to follow.
I'm not sure how much TSMC values the non-disclosure about the manufacturing node that Avalon used. But before they had their chips manufactured by TSMC they had to sign something about obeying reasonable care to avoid disclosing TSMC-proprietary and whoever-else-proprietary information. SHA-2 is an example of a self-testing structure, something akin to the test structures used in the manufacturing process testing and calibration.
When Avalon is going to disclose their voltage regulator and clock synthesizer programming information it will allow competent people to obtain very detailed information about TSMC process used. I don't think that theres much commercial value in that, but it is the intentions that count. Avalon signed not to disclose, but allowed disclosure through carelessness. TSMC aren't going to be thrilled about it and will drive harder bargain when Avalon tries to order the 2nd batch. I'm not expecting somebody from Chronicle Technology open the account to post "Thanks, suckers.", but maybe some of the Avalon competitors will do that.
The technique I'm talking above is called "yield estimation using test data". SHA-2 is 100% self-testing and trivial to reverse-engineer. Competent semiconductor manufacture engineer can with the help of changing clock frequency and supply voltage obtain a highly proprietary data in a completely non-destructive way (no chip desoldering, decaping, etc.)
An observant reader may ask "why neither Intel nor AMD seem to care about chip with unlocked clock-multiplier and voltage identifier". The answer is "binning". A large manufacturer will do an extensive test of their chips and sort them into bins. When they sell the "enthusiast-grade" chips with unlocked clock they sell them from the "fastest process corner" bin. All other bins are clock-locked and sold cheaper into OEM market. By "bin sorting" the manufacturer can completely obfuscate actual process parameters and make competitive yield estimation pointless.
On the other hand Bitcoin ASIC vendors cannot afford detailed chip testing, both because of financial and time constraints. Any chip that passes quick needle-test on the wafer prober will be packaged and mounted in the shipping product. This situation gives the analyst the sampling of an entire defect/yield curve for the fabrication process.
As far as I know the most commercially/competetively interesting information is obtained by testing the worst chips, those from the bin nearest to the "trash bin", the ones that barely met the specifications. In fact the content of the "trash bin" is quite valuable to the competition and therefore each fab carefully destroys the chips that failed the acceptance tests.
I hope that the above addition my posts will look more like grey magic than black magic. Please do some web searches about "yield estimation chip" and read the freely-available information.
The process that seems to be used is quite old. There is not a lot anyone can learn on this scale. Competitive electronics design is happening on the extremes where new structural designs are actually
needed for progress.
At the level avalon delivered they were working with well developed tech using macro blocks.
I'd guess not very interesting to the competition.
Moreover, avalon could have easily developed cgminer in a way that would separate the TSMC sensitive data.
While you claim that there
must be sensitive data in their version of cgminer i see no proof of that. In fact, because they don't release the code (even in redacted form) there is no way to actually check. So your claims are
completely baseless.
But in any case, if there actually IS sensitive data baked into this version of cgminer then avalon have put themselfs at odds with either the fab license or the GPL. One requires them to disclose what the other forbids to disclose.
This is a situation that is entirely avalons fault and should not have existed in the first place.
That alone should be enough for the community to shunt avalon as they turned out to be greedy bastards.
