Yes, in vanilla TM, a single winner per minicolumn is chosen in the previous time step, and if a new segment needs to be grown in the current time step, it connects to just the winners.
Note that not all TM implementations follow this, though (for example, etaler which I mentioned on one of your earlier threads will grow connections randomly to active cells in the previous time step, not just the “winners”).
The vanilla TM algorithm will:
- Only grow connections to previously active winner cells (not to ones which were not winners)
- If a segment has “max new synapse count” active synapses when its post-synaptic cell becomes predicted active, then do not grow any additional synapses (to winners or otherwise)