how many corner cases to sure the circuit being mass produce

[purefen]

snsp corner

Dears:
Although a designer should know which corner case will degrade the performance of your designed circuit. But it is maybe the worst case which is out of 3-sgma ramge. If trying to overcome the worst case, the area and power of the circuit will raise. Will you do that? How many corner cases to sure the circuit being mass produce?
corner case:
rratio, cratio, i_ratio(internal current), ixp_ratio(external currnet), mos(tt,ff,fs,....)......
BR. purefen

 

[micel99]

As I understand, it depends on how much worse you expect the process could be.

 

[eeweiboy]

Re: how many corner cases to sure the circuit being mass pro

some better FAB controll their process in TT corner

 

[survivor]

Re: how many corner cases to sure the circuit being mass pro

Hi,

From my experience, i usually use the folloing condition ;
(PS: Take 0.35um process as examples for consumer product)
1. VCC : 2.9V, 3.3V, 3.7V
2. TEMP : -20C, 40C, 100C
3. Process corner: Typ, FNFP, SNSP, FNSP, SNFP
4. Resistor value: 70% , 100%, 130%
5. Cap value : 80%, 100%, 120%

So the total condition= 3x3x5x3x3 = 405 condition.

 

[Humungus]

Re: how many corner cases to sure the circuit being mass pro

Survivor is in the way, but corner analysis give quite pessimistic results. That means that you must oversize your design (are and power consumption) to satisfy that your circuit works well in every corner (as survivor said: a lot).

Normally, even if this is more time consuming, Monte Carlo analysis, if you have some correlations defined between the corners of your devices, give more realistic results. AMS has written some presentations showing another method called statistical corner, which takes into consideration that corners parameter of your devices are not independent of each other. This results in a correlation between parameters, reducing so the number and modifying the "form" of the space of corners.

See attachements

 

[purefen]

Re: how many corner cases to sure the circuit being mass pro

Dears:
Thanks for your replies. Read all the replies of your, I thinsk that Humungus's suggestion is more realitic and logical. The more corner is, the more tired the designer is. As the result, I will try to find out the flow that Humungus mention and verify it. Thank you a lot.
BR. purefen

 

[fashion]

I think the more the better,but its will influence the time to market,so at least 5 cornners.

 

[electronrancher]

405 corners? no, designer's job is to produce circuits, not sim 7-transistor op amp for 6 weeks.

my opinion is that fs/sf are rare, and i use them to find out which flavor of transistor is misbehaving when a circuit falls off a cliff in ff/ss.

the reason fs/sf are rare is because main factor in fastness is tox, which is about equal for all mosfets in a given area.

actually, i try not to depend on corners for figuring out what i should already know - for example let's think about an op amp. main specs are gain and phase. gain is worst at hot/slow, so satisfy gain using hot/slow. phase margin is worst at cold/fast, so use cold/fast to satisfy phase margin.

only if you don't understand a circuit should you be running all the corners (meaning fs sf). and then you should be running this corner over temp to look for the delta.

ie - cold makes fast, hot makes slow, so -40, +140 at ff lets you see how each device acts as it slows down in it's fastest zone, you get a slope within a slope. understand that, and you'll crank out 10 chips in the time that others run 405 corners on a single block..

ps - typ typ is a given. you should probably draw in typ typ initally, but i wouldn't spend a whole lotta time running voltage and temp corners on typ typ if your circuit passes those corners at fast & slow!.

just my opinion, and this time i won't even charge you the two cents i usually do.

 

[CDRCDR]

I simulate the SS first

 

[ecijun]

thanks Humungus

 

[leo_o2]

Re: how many corner cases to sure the circuit being mass pro

You can employ Monte Carlo analysis for it.
For example, .TRAN 1n 10n SWEEP MONTE=val
You can use 30 for val to achieve a 99% possibility from normal 80% possibility.

 

[selvaraja]

Re: how many corner cases to sure the circuit being mass pro

1.Run simulation all the process corner Typical, FNFP (fast NMOS fast PMOS, SNSP
(slow NMOS slow PMOS), FNSP ( fast NMOS slow PMOS , SNFP ( slow NMOS fast PMOS ).

2.Simulate for +/- 10 % from supply voltage

3. temperature -40 to 120 degrees.

your design is pass if meet at least 70-80% of all the corner parameter

 

[nus_lin]

Re: how many corner cases to sure the circuit being mass pro

that is great, i found what i am looking for. (sorry, i need to get some point to download file)

 

[andy2000a]

if circuit include bandgap or use parasitic BJT cell
but many model only Typ BJT model , how to simulation it ?

 

[ambreesh]

Re: how many corner cases to sure the circuit being mass pro

Dear Andy2000a,
Change the IS vaue given in the model for BJT by +/- 30%.

 

[Teddy]

Re: how many corner cases to sure the circuit being mass pro

It all depends if your management understand what you are doing - are technical or not.
If not you'll end up simulating all corners and MonteCarlo.
Have fun

You should also consider that best cases (+3sigma) do not happen. But -3sigma does happen relatively often (in a large scale of things) so you should run WS. MonteCarlo is more realistic for sure.
It also depends if you can live with idea of scrapping a lot if it is WS one.