I. Sharovar
Truememorytechnology LLC,
United States
Keywords: storage for embedded systems, hardware for AI, NVDIMM, power efficiency
Summary:
The proposed technology develops high-performance and low-power storage devices for embedded systems. The storage devices would extend the current market and increase the usage of high-performance applications, including Artificial Intelligence (AI), in small and portable configurations. Embedded systems now use SD cards or PCIe NVMe SSD as storage devices. SD cards consume low power but don’t provide enough bandwidth for high-performance applications. Even the new SD Express standard, which uses a PCIe interface, still could not be close to high-end SSDs in terms of performance. NVMe SSDs provide high performance but have higher power consumption and CPU utilization, making them difficult to use in small, portable configurations. However, the new type of embedded applications requires storage with high bandwidth. Examples of such devices are Edge devices that work as network endpoints and process data before sending it to the Cloud. They often execute high-performance applications and require storage with small size and low power. At the same time, such applications need bandwidth close to the server type of SSDs. Because the current embedded storage devices cannot satisfy these requirements, they have become the main pain point of customers. The proposed technology designs devices with small sizes, low power, and bandwidth similar to high-end SSDs. For particular types of applications, the speed of storing data could even exceed the speeds of high-performance SSDs currently on the market. For example, an HD video camera, using the technology, could store data in real-time with a higher resolution than would do it now with NVMe SSDs. In addition, the proposed technology would develop a communication interface that extends the usage of AI hardware solutions in Edge devices. Now, hardware accelerators, including AI chips, use a PCIe interface to communicate with host applications. It limits the use of such hardware in small devices. Our technology would eliminate the PCIe interface and allow the use of high-performance AI applications in a bigger range of Edge devices, especially in portable and compact configurations. It would increase the size of the targeted market. The proposed technology will design a memory device that would work on existing DDR4/DDR5 memory buses using standard memory controllers. The memory module would be designed either as on-board memory or as an NVDIMM module. It would use one memory interface for both volatile (RAM) and non-volatile memories and could eliminate the PCI interface, decreasing power consumption and the physical size of embedded devices. There have yet to be similar storage devices in the embedded systems market. The technology could not only compete in the existing market but would also help create new storage device types for high-performance applications used in small systems. The technical ideas used in the technology are innovative and protected by the granted patents.