"I have the below requirement for server switches of 10 switches,How can I size the core switch "
You can size the core like any other switch, i.e. how much bandwidth, and PPS, are expected to pass through it. Further, assuming higher bandwidth ingress to lower egress bandwidth, you might take into consideration what buffer resources you'll need. Also, as some switch's internal architecture has bottlenecks (a bank of ports on the same ASIC), not apparent from overall performance stats, you often need to account for those too.
Although you've noted you'll have 10 switches with, at least (???), 160 Gbps fabric and, at least, 100 Mpps forwarding capability, this tells us almost nothing what those switches will pass up to the core and/or bandwidth link(s) to the core.
"What are the criteria for deciding switch fabric and forwarding rate for an access switch ,TOR ,core switches "
Again, having an expectation for volume/kind of traffic, and where to/from.
Many Enterprise switches are now noted as being wire-rate (for even minimal size Ethernet frames) on all ports concurrently. I.e. in theory, every switch port can be running at 100%, concurrently, without dropping frames/packets. However, the moment you have just two (or more) ports send to one port, you can congest the egress port, or, likewise, if you have something like an 100/10 Gbps port sending to a 10/1 Gbps port (respectively), you can congest the egress port. Fabric bandwidth and/or PPS don't help much then.
Typically, we have over-subscription from the edge of the network toward the core of a network. We also have old "rules-of-thumb" for typical, acceptable, over-subscription ratios, such as 24:1 for host edge switch link uplinks, 8:1 for server edge switch uplinks, (perhaps) 4:1 for distribution uplinks. However, such ratios may be far off for your service needs.
Personally, in general, with the leaps and bounds in bandwidths, it seems, to me, for some situations, the aforementioned ratios are two generous, or even, insufficient. For example, when hosts edge ports migrated from 10 to 100 to gig ports, "typical" host traffic didn't also, concurrently, jump 10x for each edge port bandwidth upgrade. Conversely, though, as modern servers might be hosting multiple "logical" (i.e. VM) servers, those server edge ports might have more traffic than individual servers had in the past.
BTW, is your question for some kind of student exercise? If so, or if not, how were the 160 Gbps fabric and 100 Mpps requirements, determined?
