Test-bench for hysteresis window of a StrongArm Latch comparator

Dear Thais,

Thais said:

Fig3 - Book's output

Question 1: How can this method be reliable if Vh varies according to how many steps I set in the parametrics analysis? 

I think the issue of quantifying the hysteresis has far more variables than the number of steps in your set of transient analyses. Let me pass a few reasons for my comment along for your consideration.

1. The response time of any comparator is a function of its gain - independent of the hysteresis issue. As a result, the propagation delay time will also vary depending on the gain and the slope of the input signal as it crosses the input threshold. As such, if you measure the value of vi1 for two input ramps with different slopes, you will measure two values for vi1 since the time it takes for the comparator to respond will depend on the slope of the ramp.

2. Secondly, it is very rare for the input signal to a comparator in its actual operating environment to be a slowly rising or fall ramp. Based on [1], it follows that the actual hysteresis in operation will differ from that measured using a set of slowly rising and falling ramps. 

3. A variable missing from the analysis as you describe it is the impact of feedthrough. An input signal that rises or falls quickly or a latch signal that has significant transition times can lead to charge injection that may alter the switching threshold and hysteresis measurement. This effect suggests that a more accurate means of characterizing the hysteresis is using a latch signal representative of your actual latch signal and operating the latch at a rate consistent with its expected rate. For example, if your latch frequency is 1 GHz and you are doing your measurements with only one latch edge per transient simulation, you are missing any impact of the prior or next latch signal on the response of the comparator.

4. Items [1] through [3] basically emphasize what you have  already observed - that the amount of hysteresis ("Vh")  is not a "constant" value but is dependent on the nature of the input signal, the latch frequency, and the attributes of the comparator (propagation time, internal coupling capacitances, gain...).

5. I also noticed the "input ramp method" appears to only use a ramp with a positive slope. I personally do not think this is realistic as the response time and hysteresis will depend on the direction in which the input signal approaches the threshold as well as the comparator output state.

So, with all these words, what might I suggest? Perhaps a more reasonable methodology is to apply an input signal, perhaps a square wave, triangular wave, or even a sinusoid whose amplitude Ao is small and is centered just below, but near the comparator's threshold, say Vth -delta_V. The reason you are designing hysteresis into the comparator is to prevent it from oscillating ("chattering") if the input signal is close to its switching threshold. Hence, if you run a set of transient simulations with an input amplitude of the signal of Ao and vary delta_V from, say -50 mV to +50 mV and observe the comparator output, you can determine if the hysteresis is less than or equal to Ao if no chattering occurs when the output switches. Presumably, for one or more values of delta_V the comparator will switch. You can then repeat the simulation for a larger value of Ao if no chattering occurs. The amplitude Ao at which you start to see chattering suggest you have a good estimate to the hysteresis as +/-Ao.

The advantage to this method is it better emulates the condition in which your comparator is used and includes the impact of both a positive and negative sloped input.

I hope these few thoughts that cam to minder somewhat helpful - or at least provide you some insight into your objective Thais.

Shawn