Need +/-1C accurate on chip temperature measurement

[PSG]

Hi,

I'm looking for a technique to measure accurately the chip temperature to +/-1C.
There's obviously the deltaVBE technique (PTAT current into a resistor) but I've been told I wouldn't get the accuracy needed. I'm presently setting up the test bench to see how much accuracy it would give me, as I really don't have a clue at this point, but if someone has a ball park idea that would be helpful.
And so if that technique really is not accurate enough, what else is there out there? MOS, Schottky diode ...??

Please help.

 

[dick_freebird]

You think a simulation testbench is going to tell you
about anything but "ideal" accuracy? It's quite likely
that, outside places like AD, LTC, TI, your diodes are
not modeled well enough for sub-% bandgap or temp
sensor accuracy - neither nominal nor process-varying,
and maybe even temperature is not credible. Certainly
a "non-valued" diode in a digital flow is not to be trusted.

I'd bet on this requiring some form of trimming and/or
cal-mapping to get sub-% accuracy in production with
low fallout. That's added cost input. Are you OK with
that?

 

[crutschow]

If it's a one-time bench measurement then you can calibrate the on-chip diode forward drop over the temperature range of interest by heating the unpowered chip in an oven and measuring the chip package temperature with an accurate temperature probe. That should allow you to measure the subsequent chip temperature using the diode forward drop when the chip is operating within a degree C (you may have to momentarily unpower the chip to get the most accurate reading).

 

[PSG]

Thanks for the answers. So what you 2 are saying is that as long as I have trim capabilities, and as long as my simulation models are accurate, I can get to +/-1C.
The problem is that I have never used this process before (nor has anyone in my company) so all I can do is simulate over process, temp and supply voltage and run Monte-Carlo analysis and hope for the best.
Yet, assuming I have good models, and assuming I can trim, which sensor architecture out there would you recommend to give me the best accuracy?

 

[crutschow]

A diode-connected BJT transistor (collector shorted to base) generally has a fairly ideal diode characteristic. Perhaps that will give you sufficient accuracy. What is the purpose of this measurement? Why do you need the chip temperature to such accuracy?

 

[Analog Ground]

A diode-connected BJT transistor (collector shorted to base) generally has a fairly ideal diode characteristic. Perhaps that will give you sufficient accuracy. What is the purpose of this measurement? Why do you need the chip temperature to such accuracy?

Do you need sensitivity or actual calibrated accuracy? I would think there are large temperature gradients across most ICs and knowing the absolute temperature would have little value. In addition, you would have to control the ambient temperature, air flow, etc. closely to interpret any absolute temperature. Certainly sensitivity would be useful. Some of us would be curious what you are doing, if you could say.