- Kioxia develops a 5 TB flash memory prototype with a 64 GB/s bandwidth for edge computing
- The prototype memory module demonstrates the design of the margaritas chain keeping the bandwidth to higher capabilities
- The new module uses the PAM4 signaling and the controller controller to improve flash performance
Kioxia has developed a new type of flash memory prototype that offers great capacity and high bandwidth.
The module was created in the R&D project for the improvement of information infrastructure and communication information after 5G of Japan directed by Nento, the National Research Agency.
The module offers 5 TB of storage and a transfer speed of 64 GB/s, placing it as an alternative or potential supplement for conventional drama in data processing. Kioxia said the design addresses long -standing compensation between the bandwidth and the capacity found in the DRAM modules when organizing flash memory chips in a margaritas chain instead of a bus connection.
Flash accounts
Each flash unit, described as an account, is connected in sequence with its own controller. The design of the Margaritas chain allows the capacity to expand without degrading bandwidth, something that has been a limitation in previous architectures.
The company also developed high -speed transceptions capable of 128 Gbps yield. These use the PAM4 modulation, which encodes two signal information bits, doubling the effective bandwidth compared to traditional binary signaling.
Along with this, the new features of the controller, such as extension, help reduce reading latency, while low amplitude signaling and distortion correction increase the speed of the memory interface to 4.0 Gbps.
Kioxia reported that the prototype consumes less than 40 W when it operates at a complete bandwidth through a PCIE 6 eight -lane PCIe connection.
The combination of energy efficiency and performance is being presented as a step to make flash more practical in intensive memory roles generally occupied by dram.
The work is aimed at servers in mobile edge computing environments, where 5G and 6G networks create the need to process information closer to final devices.
IoT and real -time workloads are promoting interest in memory systems that balance the cost, capacity and performance.
Kioxia said he plans to continue developing this technology for marketing, with potential uses in IoT, Big Data analysis and advanced AI models, including generative AI.
