Research Bits: July 18

CXL memory disaggregation

Researchers from the Korea Advanced Institute of Science and Technology (KAIST) developed a Compute Express Link (CXL) solution for directly accessible, high-performance memory disaggregation that they say significantly improves performance compared to existing remote direct memory access (RDMA)-based memory disaggregation.

RDMA enables a host to directly access another host’s memory via the InfiniBand data center network protocol. However, the researchers pointed out that this method requires an additional CPU, and redundant data copies and software fabric interventions for RDMA-based memory disaggregation cause longer access latency.

The new CXL-based memory disaggregation framework includes a CXL-enabled customized CPUs, CXL devices, CXL switches, and CXL-aware operating system modules. The team’s CXL device is a pure passive and directly accessible memory node that contains multiple DRAM DIMMs and a CXL memory controller. Since the CXL memory controller supports the memory in the CXL device, a host can utilize the memory node without processor or software intervention.

The CXL switch enables scaling out a host’s memory capacity by hierarchically connecting multiple CXL devices to the CXL switch allowing more than hundreds of devices. Atop the switches and devices, the CXL-enabled operating system removes redundant data copy and protocol conversion exhibited by conventional RDMA, which can significantly decrease access latency to the memory nodes.

“Escaping from the conventional RDMA-based memory disaggregation, our CXL-based memory disaggregation framework can provide high scalability and performance for diverse datacenters and cloud service infrastructures,” said Myoungsoo Jung, a professor in the Computer Architecture and Memory Systems Laboratory at KAIST.

In a test comparing loading 64B (cacheline) data from memory pooling devices, CXL-based memory disaggregation showed 8.2 times higher data load performance than RDMA-based memory disaggregation and even similar performance to local DRAM memory. In the team’s evaluations for a big data benchmark such as a machine learning-based test, CXL-based memory disaggregation technology also showed a maximum of 3.7 times higher performance than prior RDMA-based memory disaggregation technologies.

“Our CXL-based memory disaggregation research will bring about a new paradigm for memory solutions that will lead the era of big data,” Jung added.

Optoelectronic logic gates

Researchers from the Korea Institute of Science and Technology (KIST) and Gwangju Institute of Science and Technology developed ultra-high-speed, high-efficiency optoelectronic logic gates (OELGs) using organic-inorganic perovskite photodiodes.

The optoelectronic logic gate uses light as an input signal, which results in low energy loss, and can operate only with light energy without an electrical power supply. In the device, two layers of perovskite thin films are vertically stacked like a sandwich. Binary logic operation is possible by inputting two lights of different wavelengths and intensities.

The perovskite optoelectronic logic gate can freely change the photocurrent polarity using light, making it possible to execute more than one logic gate operation result for the same input value.

This means that compared with the existing logic gate that can only perform one logical operation on one device, the newly developed one can implement all five different basic logic operations: AND, OR, NAND, NOR, and NOT. The team said that this enables the development of optical processors with high spatial efficiency and integration, as one logic gate can function like five logic gates.

“Perovskite optoelectronic logic gates that execute multiple logic operations in response to optical input are expected to be used for ultra-small and low-power universal optical sensor platforms in the future,” said Yusin Pak of the Sensor System Research Center at KIST. The researchers expect it to be useful for applications such as next-generation optical communication, optical network, and healthcare.

All-optical random bit generator

Researchers from Taiyuan University of Technology, Guangdong University of Technology, Northwestern Polytechnical University, Institute of Southwestern Communication, and Bangor University propose a method for real-time generation of physical random bits by combining broadband photonic entropy sources with all-optical signal-processing techniques.

The researchers noted that optical chaos is a reliable way to generate fast and real-time random bits, due to its high bandwidth and large amplitude fluctuations. However, most optical chaos-based random bit generators perform their quantization in the electrical domain using electrical analog-to-digital converters, creating a bottleneck.

In the team’s all-optical method for generating random bits, chaotic pulses are quantized into a physical random bit stream in the optical domain by means of a length of highly nonlinear fiber. In the proof-of-concept experiment, they successfully generated a 10 Gb/s random bit stream in a single channel.

The team notes that the current rate-time of 10 Gb/s is only limited by the adopted chaos bandwidth. Their scheme can operate potentially at much higher rates than 100 Gb/s if the bandwidth of the chaotic entropy source is sufficient.