The simplest way is to use a hot start enzyme. Either a commercial hot start enzyme or by using a cold start enzyme but adding it to a hot pcr reagent mix and starting cycling before the mix can cool down ( only for small numbers of pcrs really) . You can also try using less primer.
It can be avoided in the first place by designing primers using Primer3 or similar online software which detects dimerisation and chooses the best primer pair
Try reducing your primer concentration and see if that helps. Primer concentration should be appropriate or else it ends ups forming primer-dimers after reaction.
Pre-heat your thermal cycler to 95 degree C and then load your samples. Reduce your primer concentration. Reduce your annealing step duration. Use gradient function of our thermal cycler to obtain the optimum annealing temperature. Limit to 35-40 PCR cycles to prevent unnecessary non-specific amplifications.