Precision of hysterisis in comparator ?

[avt]

I have implemented a comparator like the on in Gregorian, "Introductionto CMOS-OP-Amps and comparators" - this means on NMOS diffpair, one cross-coupled PMOS pair, and on PMOS load transistor in each branch of the NMOS diffpair.

It works quite good under typical mean conditions with 25 mV hysteresis - but when I do corner analysis the hysteresis threshold values changes signifcantly and can be higher than 100mV. As I have a spec that demands a threshold value of at least 25 mV and an upper limit of less than 100 mV - I would like to have better control of the hysteresis threshold.

Are there any passive or active measures to control the hysteresis threshold in regnerative (positive feedback) comparators - or are there better structures than the one I use with respect to hysteresis threshold ?

 

[leo_o2]

Use two voltage references with 50mV difference and employ the comparator output to select.

 

[avt]

sorry - didn't get that - could you elaborate your design-idea a little bit more ???

just to be clear - the hysteresis is set mainly (in handcalculation) by the W/L-ratio of the two PMOS-pairs and the gm of the NMOS pair ... (the input signal is differential - so every "input" is used - so where/how to apply an external reference ?)

Added after 5 hours 21 minutes:

I mean - what would be realy useful is if it qould be possible to trim the input hysteresis with bias currents or something simlar - the question would be then : how to measure the hysteresis threshold voltages (if one assumes that they are symmetric ) for this kind of "control-feedback-circuit" ??? (I mean it would be simlar in a certain sense to a cmfb - but here we need to figure out the current hysteresis threshold ...)

 

[leo_o2]

My suggestion is avoid to create the hysteresis only by comparator. The idea is to change the voltage reference. Of course it is easier to implement if one of inputs of your comparator is fixed. If both inputs of comparator are changing, this idea can also be realized but a little more complex.