Samsung Announces New Low-Power, High-Bandwidth DRAM

This week, Samsung released a news release titled "Editorial" discussing how its memory products will "leverage the AI era." This editorial presents many broad and unverified claims about AI, such as "at home, it is making our lives easier and more enjoyable." However, the exciting part lists its developing memory products that will enable AI models to run on devices. This will be a significant challenge for the industry as these models typically run in the cloud due to their scale and processing requirements. It reveals a previously unheard-of type of memory called Low Latency Wide I/O (LLW) DRAM, according to Samsung.

This editorial does not disclose much information about this mysterious new type of memory. It simply states that this memory provides bandwidth of up to 128GB/s, equivalent to a DDR5-8000 module, as noted by Tom's Hardware. However, it achieves this throughput with only 1.2pJ/b energy consumption, which is relatively low, suggesting it may be targeted towards smartphones and laptops rather than desktop PCs. There is also no indication of the speed at which the module achieves high bandwidth and low power consumption. We need to wait for more detailed information to understand which products it can be used for and whether it will be strictly for enterprise or client applications.

Undoubtedly, the rise of AI will bring significant challenges to companies like Samsung, assuming these AI models will migrate from the cloud to our devices. AMD, Intel, and NVIDIA have already begun touting the benefits of using dedicated hardware (such as Neural Processing Units (NPUs) and Tensor Cores) to run AI applications on devices. However, there is a lack of applications running and efficient execution capabilities. Solutions like LLW DRAM may be part of the technology family supporting the transition of AI models from the cloud to devices. Nevertheless, like with these things, we will have to wait and see if this technology has practical implications or if it will become a new metaverse by the end of this year.