Samsung, SK Hynix partner on processing-in-memory
Memory chip manufacturers Samsung and SK Hynix are collaborating to standardize processing-in-memory (PIM) in the form of an LP-DDR6 memory, according to Business Korea.
The collaboration includes preliminary work to register the standardization of the LP-DDR6-PIM with the Joint Electron Device Engineering Council (JEDEC).
The two companies are rivals across the memory sector and particularly in the supply of high-bandwidth memory (HBM) format chips which are used extensively in artificial intelligence. However, they are putting aside their differences to commercialize a next-generation memory that could change AI processing architectures.
Already in the market
Samsung claims it is the first company to integrate Processing-in-Memory (PIM) in HBM DRAMs. PIM is able to process some logic functions close to memory by integrating compute engines on the memory die. These functions avoid data movements and could reduce power consumption in AI applications. Samsung claims a 2x improvement in performance and a 70 percent reduction in power consumption is possible compared with the use of conventional HBM. The power savings from PIM are considered particularly significant for on-device AI in smartphones and personal computers.
However, the successful introduction of PIM depends on how functionality is partitioned and achieving consensus across industry or at least amongst memory suppliers.
Samsung Electronics introduced HBM and LPDDR5 products with PIM and pursued standardization, but it did not come to fruition, Business Korea reports. SK Hynix also released a Graphics DRAM (GDDR6) with PIM.
“The two companies are exchanging opinions and cooperating for product standardization,” Business Korea quoted an unnamed Samsung representative saying. “Since the cooperation has just begun, we are in the process of setting the implementation plans for the standardization target timeframe.”
Related links and articles:
News articles:
Samsung presents AI processing-in-memory options
Samsung embeds AI accelerator in memory chip
Unsupervised in-memory AI learning goes digital
Processor-in-memory DRAM benchmarked on Xeon server
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
