The Right Memory for the Right Market at the Right Time

Potential applications for 4DS’ ReRAM technology

We see several strong potential applications that can take advantage of the unique features of the 4DS cell, including  Warm Data storage, Persistent Data Backup, and Analog In-Memory Compute.

In Warm Data storage, the 4DS cell would be a complement to DRAM but not replace it. DRAM will always need to do the heavy lifting of immediate data access and storage for data intensive processing. However, in emerging applications there is growing awareness that not all data is the same, that you can segment data based on how often it changes, such as in AI Inference Engines which need billions of weighted parameters at GPU speeds but whose values are static or slow-changing. At the same time you can’t sacrifice access speed for partitioning or data retention without impacting system performance or scalability. This is exactly where 4DS’ write speeds, persistence and scalability can fit in.

In Persistent Data Backup a technology that can provide the speeds of DRAM with the persistence of Non-Volatile can enable faster data access, improved system reboot in the case of fault, extra memory capacity to DRAM for big data analytics, or can be ultrafast storage for in-memory databases. The exponentially growing scale of AI clusters brings with it increased risk for system faults and significant downtime for system recovery, so an always-available high bandwidth persistent memory solution can provide a critical backup to these increasingly mission critical systems. 4DS believes there is still an unmet need for a high bandwidth persistent memory solution and that 4DS could potentially deliver that solution.

The third potential application is in emerging AI processor architectures that look to take advantage of 4DS’ analog characteristics to develop more energy and processor efficient solutions for neural net processors, an approach called Analog In-Memory Compute. This is becoming more important to the overall industry as AI processing is forecasted to consume 10x as much energy by 2026 as it did in 2023.