Verify Patron Calculated with airSlate SignNow

Verify patron calculated

hi I'm Esther Lee I'm the editor-in-chief of semiconductor engineering I'm over at synopsis with JT longjin oh it's going to talk today about formal data path verification so JT what is formal data path verification what defines that versus other kinds of verification so when most people think about formal verification they think about the control path they think about the the pipelining and the state machine elements at all that different parts of your design when we talk about the formal data path now we're looking at leveraging formal solvers and formal methods to the data manipulation or data transformation areas of your designs rather than the control path areas so why is this becoming a big problem now versus what it was in the past well data path verification has historically been a hard thing in our industry and there have been a number of well-publicized occurrences that incorrect data manipulations have had significant financial or even physical losses with the rise of data paths heavy IP like in machine learning and artificial intelligence the requirement to have high confidence and your data path designs above what you can get with simulation has become all the more important why don't you draw this out for us sure what are we looking at here so I mentioned before that simulation often times is insufficient to get the kind of confidence we want and our data path designs and this is an example of that this is a two operand multiplier block something that most people have had familiarity with or they've seen before this multiplier block might take 16-bit operands 32-bit operands or 64-bit operands now if we were going to try to prove this block exhaustively what kind of time might we expect it to take well let's go beyond anything we can get with our simulators anything we can get with our emulators and try to run it exhaustively on a commercial piece of processor silicon and say we can retire three billion operand pairs every second time is a major consideration here right well of course if you have 16-bit operands you have two to the 32 operand pairs and if you can retire three billion operand pairs a second this will take you about one and a half seconds no problem we're doing great now let's say we have 32-bit operands this takes us to two to the 64 operand pairs and at 3 billion operation pairs a second now it's going to take us roughly a hundred and ninety five compute years to go through all of these operand pairs that is possible maybe if it's something that's really really important and we absolutely have to have it exhaustively correct but it's unlikely that anyone's going to want to spend to take that much verification resources towards one function the way this has been done in the past is pretty much a brute-force type of approach right what you're doing is adding a...