Yes, TLBs are pretty expensive (in area and power) since they are associative memories. The virtual (non-offset) part of an address to be translated is compared

Yes, TLBs are pretty expensive (in area and power) since they are associative memories. The virtual (non-offset) part of an address to be translated is compared to all TLB entries simultaneously to keep latencies down since the TLB is on the critical path from CPU to memory. Accordingly, you need one comparator (lots of XOR gates in a simple implementation) per TLB entry.