For devices for which only min and max propagation delays are specified, one might safely -- or optimistically -- assume that best case typical could be somewhere in the middle of that range. But what about devices that only specify maximum? In RAMs, for example, how likely is it that (as long as layout and construction isn't overly sloppy,) tACC or other parameters could be less than the max value stated?

I wouldn't suggest that such an assumption would be valid for every device in every circumstance, but as one example of performance exceeding datasheet specs, we've seen that the 65C02 can be clocked faster than what WDC specifies.

How much Truthiness, one way or the other, can we count on?

_________________
"I am endeavoring, ma'am, to create a mnemonic memory circuit... using stone knives and bearskins." -- Spock to Edith Keeler

_________________
http://WilsonMinesCo.com/ lots of 6502 resources
The "second front page" is http://wilsonminesco.com/links.html .
What's an additional VIA among friends, anyhow?

Also, note that the W65C02 and W65C816 chips are a particular case of the chips now being produced with a process that is faster than the process of the chips that the specification testing was done for. It seems like it's just not worth it to WDC to redo the testing.

But from an account that I read of the days that CMD was producing the SuperCPU accelerator for the Commodore 64 by W65C816-14 chips at 20MHz, they were testing every single chip they bought and discarding a large number of them as unsuitable. So if 99% of the chips produced with the original process were good for 14MHz, perhaps only 50%-60% were good for 20MHz ... and even with that testing, the SuperCPU accelerators had a reputation back in the day for being temperamental (which as the thread Garth linked to discusses, is sometimes a temperature issue itself).

With the process they are making them with now, it seems likely a lot more would have passed that testing.

My project hasn’t missed a beat at 25 MHz, so I’d say there is some leeway. This is with both the 02 and the 816.

_________________
x86?  We ain't got no x86.  We don't NEED no stinking x86!

Yeah, those 1.0µ '816's are the ones that CMD were fighting to get about half running at 20Mhz in their SuperCPU. I forget who I heard that from ... maybe Jim Brain at Retro Innovations?


I couldn't come close to affording one of those back when they came out, but I would have loved to have one.

I just assume 0.6u W65C02 will run at 25Mhz, every one of them. I know some will run to mid-30Mhz.
Bill

Note for this that 6T in a W65C02S6TPG-14 refers to a "0.6u 1P/2M CMOS Process" at the TSMC Foundry (6TPLG-14 for PLCC and 6TQG-14 for QFP).

The older W65C816 that I was referring to would have a different process number.

_________________
x86?  We ain't got no x86.  We don't NEED no stinking x86!

There are lies, damn lies, then statistics, and finally data sheets. Most parts will generously exceed their "specifications".

One of my favorite "specifications" has to do with power supplies. Every one will have the noise specified at a 20MHz bandwidth (with added filtering to boot!). It might be belting out several Vpp at 100MHz while quiet as a mouse at 20MHz and below. Why? Because that's the way it's always been done since the best scopes in the world were only capable of 20MHz and no engineer will ever mess with tradition. Even if such a spec is useless for the intended use.

_________________
Bill

Spec is validated at temperature, voltage, and process extremes. It also has test load capacitance and resistance that maybe overly conservative. W65C02 also had significant die shrink without changing timing spec, so a W65C02 design with few components to reduce loadings, operates in nominal voltage and temperature can really push the spec. The biggest change was die shrink to 0.6u. I understand they wanted the benefits of smaller die and greater yield, but don’t want to change the test vectors, so it is up to us hobbyists to find the upper limit of W65C02 as well as W65C22 that also had the die shrink process.
Bill

You all got fixated on the 2nd part of my query. My bad, I guess, for citing the 65C02 as an example of performance exceeding specs. What about the first part of what I asked, about logic and other devices for which only min and max prop delays are specified, omitting typical, or memory datasheets that only specify max tAA? A 25nS part might be faster than that because it's one that didn't make the grade as 15nS because its tAA was17-18nS.

_________________
"I am endeavoring, ma'am, to create a mnemonic memory circuit... using stone knives and bearskins." -- Spock to Edith Keeler

_________________
http://WilsonMinesCo.com/ lots of 6502 resources
The "second front page" is http://wilsonminesco.com/links.html .
What's an additional VIA among friends, anyhow?

Agreed. This is how they categorise modern CPU’s like Intel and AMD’s, for example.

Using Pentiums as an example back when the only difference was clock speed, it was exactly that. They all came from the same fab but were classed on what tests they passed. You might get a bunch of Pentium 200’s out of a batch (that passed the 200MHz tests), then the ones that failed at 200 but were stable at 166, 133 MHz etc, were stamped accordingly and sold as Pentium 166’s and so on.

Comes back to the old “silicon lottery”.