Ultracompact photonic integrated content addressable memory using phase change materials

The inherently parallel and ultrafast nature of photonic circuits is naturally well suited for implementations of content addressable memory (CAM) circuits which perform highly parallel, in-memory computations leading to extremely high throughput compare operations at low latency in applications such as network routing, associative learning and CPU caching. Towards improving the latency by orders of magnitude over conventional electronic implementations, we propose a scalable photonic architecture for a CAM based on silicon microring resonators embedded with patches of phase change material Ge2Sb2Te5 (GST-225). The CAM cells behave as frequency selective 1 x 2 switches and emulate the digital XOR operation in a wavelength division multiplexing (WDM) read scheme at speeds limited only by the bit modulation system and photodetector response. Through device and circuit level simulations we show that the proposed reconfigurable nonvolatile architecture can reliably perform tens of Gbps speed read operations in the presence of channel noise, crosstalk and source-receiver nonidealities. The individual CAM cells have a compact footprint of less than 10 mu m(2) and require no continuous power input to maintain the switch states, which is naturally advantageous given the scalability requirement and long intervals between entry updates in modern CAM systems.

Automated summary

The inherently parallel and ultrafast nature of photonic circuits make them well-suited for content addressable memory (CAM) circuits, which can perform highly parallel computations with high throughput and low latency. This paper proposes a scalable photonic CAM architecture using silicon microring resonators embedded with phase change material patches, offering reconfigurable nonvolatile capabilities and reliable high-speed read operations in the presence of noise and nonidealities.