Welcome Here  Shenzhen Mingjiada Electronics Co., Ltd.

Kioxia Corporation, a leading enterprise in the global storage solutions field, has successfully developed a prototype of a large-capacity and high-bandwidth flash memory module, which is crucial for large-scale artificial intelligence (AI) models. This achievement was made within the framework of the Post-5G information and Communication System Infrastructure Enhancement Research and Development Project (JPNP20017) commissioned by Japan's national research and development agency, the Japan New Energy Industry Technology and Development Organization (NEDO). This storage module features a large capacity of 5 terabytes (TB) and a high bandwidth of 64 gigabytes per second (GB/s).

To address the issue of the difficulty in balancing capacity and bandwidth faced by traditional DRAM storage modules, Kioxia has developed a new module configuration that adopts a daisy-chain connection method of flash memory particles. We have also developed high-speed transceiver technology that supports 128 gigabits per second (Gbps) bandwidth, as well as related technologies to enhance the performance of flash memory. These innovations have been effectively applied to storage controllers and storage modules.

The practical application of this storage module is expected to accelerate digital transformation and promote the application of Internet of Things (IoT), big data analysis and advanced AI processing in scenarios such as post-5G /6G mobile edge computing (MEC) servers.

1. R&d background

In the post-5G /6G era, wireless networks are expected to achieve higher speeds, lower latency, and the ability to connect more devices simultaneously. However, transmitting data to remote cloud servers for processing will increase the latency of the entire network, including wired networks, causing difficulties for real-time applications. Therefore, it is necessary to widely adopt MEC servers that can process data closer to users, which is expected to drive the digital transformation of multiple industries. In addition, in recent years, the demand for advanced AI applications such as generative AI has been constantly increasing. With the improvement of MEC server performance, storage modules also need to have greater capacity and higher bandwidth.

Against this backdrop, Kioxia focused on enhancing the capacity and bandwidth of flash-based storage modules in this project. The company has successfully developed a prototype of a storage module with a capacity of 5 TB and a bandwidth of 64 GB/s, and verified its operability.

2. Project achievements

2.1 Use Daisy chain connection

To achieve a storage module that combines large capacity and high bandwidth, Kioxia adopts a Daisy chain connection method, with each storage board connected to a series of controllers instead of using a bus connection. Therefore, even if the number of flash memories is increased, the bandwidth will not decrease, and it can achieve a large capacity that exceeds the traditional limitations.

2.2 128 Gbps PAM4 high-speed low-power Transceiver

The Daisy chain connection between storage controllers adopts high-speed differential serial signal transmission (rather than parallel signal transmission) to reduce the number of connections, and utilizes PAM4 (four-level Pulse Amplitude Modulation) technology to achieve a higher bandwidth of 128 Gbps at low power consumption.

2.3 Technologies for Enhancing Flash Memory Performance

To reduce the read latency of flash memory in the storage module, Kioxia developed flash memory prefetching technology - by prefetching data during sequential access to minimize latency, and applied this technology to the controller. In addition, by applying low-amplitude signals and distortion correction/suppression techniques in the interface between the storage controller and the flash memory, the storage bandwidth has been increased to 4.0 Gbps.

2.4 Prototype Development of Storage Controller and Storage Module

By applying 128 Gbps PAM4 high-speed low-power transceivers and technologies that enhance flash memory performance, Kioxia has developed prototypes of storage controllers and storage modules. The host interface between them and the server adopts PCIe® 6.0 (64 Gbps, 8 channels). This prototype storage module demonstrates that it can achieve a capacity of 5TB and a bandwidth of 64GB/s with a power consumption of less than 40 watts.

3. Future plans

In addition to the Internet of Things, big data analysis and advanced AI processing at the edge, Kioxia will also rely on new market trends such as generative AI to promote the early commercialization and practical application of this research result.

For more industry news and information, visit www.hkmjd.com.