The need for higher speed supercomputing power at NASA is clear, and one of the many impairments to realizing an exaflop computer is the speed and latency of the data links that connect the processors and memory. Even with the advent of higher speed standards in Ethernet and Fibre Channel, the data throughput is crippled due to the need for non-binary encoding and error correction algorithms as a direct result of the limited bandwidth of the optical components. The induced latency and power consumption are acceptable in a data center environment but severely limits high performance computing. Efforts to increase the fundamental optical component bandwidth have not been successful in increasing to the point of achieving binary NRZ data transmission without error correction at 100Gbps. Binary NRZ is the most power and latency efficient means to transmit data in a super computing environment. A revolutionary optical component is needed to break the traditional paradigm. The laser and electro-absorptive modulator (laser-EAM) in this proposal represents many significant technical advancements including: (1) fundamental bandwidth in excess of 100GHz using direct band to band absorption, (2) very low power consumption (<150fJ/bit), (3) operation over the entire environmental temperature range without a thermo-electric controller, (4) monolithic integration of the laser and EAM (5) potential to integrate a vertical outcoupler. The innovations that lead to these technical achievements are the design of a common active region for both the laser and EAM, and the use of enhanced coupling strength (ECS) gratings to provide wideband mirrors. This combination allows the laser emission to tune freely with temperature and the EAM absorption to track the laser. Further, the laser-EAM is realized in a single epitaxial growth, and the fabrication is completely monolithic. This provides a highly manufacturable and cost-effective solution to the optical component bandwidth problem.
The primary application of the laser-EAM will be in connectivity of processors and memory banks in a high-performance computing environment. Modern computing is limited by the data latency of the links between the computing elements, and currently available high-speed optical components use non-binary data encoding and error correction algorithms to achieve the data bandwidth. The proposed laser-EAM resolves the optical component bandwidth limitation and will enable higher speed and lower power consumption in the data connections.
The global network traffic continues to grow exponentially, and optical communications standards such as Ethernet and Fiber Channel need a clear path to 100Gbps serial data connectivity to support the demand. Current solutions are power inefficient and have high data latency. To enable an exaflop computer, which is expected in 2022, more than 1M links operating at 100Gbps are needed.