Importance sampling to accelerate Monte carlo simulation

In transistor level circuits for smaller technology nodes, the on-chip process
variations causes many reliability issues. To find the probability distribution of circuit delay I would like to do
a Monte carlo simulation by sampling channel length distribution of each transistor randomly and find the circuit delay for
each sampled value of channel length. The channel length distribution is assumed to be normal.

For reducing the cost of this large number of simulations, importance sampling can be used which
makes a decision to use a random sample or not based on the weight of the sample. As far as I understand importance
sampling, I need another probability distribution B along with the distribution A (channel length normal distribution).
The weight is the ratio of two samples from each of these distributions. Am I right? If yes, how can I choose a
suitable distribution B?