Probit Analysis on Limited Data

My problem is how to address data on mortality rates of spiders exposed to three temperatures. I have done probit to analyse similar data with multiple temperatures with no problem. However, I am getting an error in my analysis of these data:

High temp (0 deg C) --- 20/20 survival
Med temp (-2 deg C) --- 6/20 survival
Low temp (-5 deg C)--- 0/20 survival

I am assuming the error is because the extremes have either no successes or no failures. Is there a legitimate way to "correct" for this problem and still do probit analysis so I can get an estimate of the lethal "dose" of temperature for 50% mortality?

One suggestions was to add one success and one failure to each of the extreme temps and then analyze using probit. The data would then look like:

High -- 21/22
Med -- 6/20
Low -- 1/22

Can anyone tell me if this is a statistically valid approach, or is there simply not much to be gained from analyzing these data?

The number to the right of the slash is the sample size -- to the left is the number of mortalities.

Mortality Survival Total (N)

High 0 20 20
Med 6 14 20
Low 20 0 20
so if you are modeling survival rate, your last cell will be empty. This makes it impossible to run an appropriate probit model. Unfortunately by adding 1 success to the high score and 1 failure to the low score, you are changing the odds dramatically. Perhaps if you had huge sample size, then adding one would not make much difference in the odds, with your sample it does.
I've similar problem.
I agree to add one success and one failure to each of the extreme values could dramatically change ld50.
What about software calculating ld50 that could process data with 3 points and with extreme values without adding extra data?
How reliable are these results? May I trust for such software?