06-05-2019 08:31 AM
I just exported the data of an FPGA to use the package flight delays in routing a PCB.
I just found something not much scientific... there is a minimum and a maximum value, in ps, numbers with up to three decimal places. But the difference between the max and min is about 1 ps.
Now, how credible it is to say that something has lower and upper limits with femtoseconds resolutions while the value itself has an unaccuracy of about 1000 fs? do the limits mean the actual values cannot go beyond them by a single femtosecond? Copying as many decimals as a calculator provides has implications.
Even if the values are correct, the way of presenting them is not. Better to give a mean value (average) and a deviation.
Thanks for bearing with this pedantic venting.
06-05-2019 09:42 AM
What family? What device? What timespec?
Generally speaking (my 15 years in IC design), one uses hspice simulations of the extracted layout for all numbers for all process corners and temperatures and voltages. They get vetted by running thousands of designs with the tools. If there is something strange, it gets caught, and looked into (number might get changed). As the designs are all well known (gathered over the years, targeted to new nodes as they appear), the vetting of the speeds files results in better quality as time goes on (before tapeout). Once first silicon arrives back, actual performance is gathered, and again, anything strange gets touched. But most numbers are automatically gathered, untouched by anyone, unobserved. Does ps resolution make sense? Perhaps not. Some values are derived by formulas, so dividing a number might easily lead to single ps values.
In regards to flight time in silicon, or in the package, these are even easier to extract automatically, and probably never get viewed or touched.
06-07-2019 12:41 AM
Nothing to do with that. Not an engineering complain but scientific. When you measure something, you take, say, ten thousand values, and they all fit between two limits you can determine with a fs accuracy then it would make sense to say "this is between 35.123 and 36.567 ps". This implies it's not possible to be 35.122 or 36.568 ps. But this is not how things are in the real World, you find values distributed around some mean and with a decreasing probability of occurring the farther away they are from that central value. A normal distribution, for example, has no such hard limits. So specifying them doesn't make sense, except under the assumption of a probability distribution and those limits representing some plus and minus unts of variance, but this just a simplification of the correct specification of a real, randomized measure: a mean and a variance measurements. That would be correct.