Ultralow-latency Interconnection Networks Have Become a Necessity in Modern High-performance Computing Systems. Recent Advances in Photonic Integration Technology Are Paving the Way for a Disruptive Step in the Design of These

......Contemporary high-performance computing (HPC) systems use the distributed shared-memory (DSM) paradigm, in which the entire memory is logically shared among all the processors but is physically implemented using memory modules distributed across many computing nodes. This approach simplifies programming, provides portability of software, and exhibits improved scalability over traditional shared-memory systems. Large-scale DSM systems, however, suffer from a fundamental communication problem that significantly affects scalability: increased latency of remote memory accesses. With faster processor speeds, the remote-access latency problem is becoming critically more pronounced, as each memory access consumes a correspondingly larger number of clock cycles. The latency of a message through a network has three contributing components: