It is indeed a year. The latencies are based on the calculations done by Colin Scott in https://github.com/colin-scott/interactive_latencies and support projecting out into the future. Sorry it's not as obvious as it could be.
The main memory one stays 100ns for every year?
It’s indeed been about 100ns for a long time. Part of this is that memory is larger though, so there may be more decisions to make to look up a line (and those are made faster). And throughput has improved. Some consumer high-end desktop hardware (think gaming rather than workstation) can have lower latency ram.