What is a 3D XPoint DIMM? – Galaxy PC Tech

You can also be interested in these:

Developed jointly by Intel and Micron Technology Inc, the 3D XPoint storage technology is a novel form of memory that has the potential to bridge the gap between RAM (DRAM) and NAND flash storage in the market. Currently, both companies are working independently to advance and market their respective products, leading to some intriguing outcomes.

How 3D XPoint memory works?

The 3D XPoint memory functions in a distinct way from other flash-based products. In 2015, Intel and Micron announced its creation, boasting its potential to be up to 1000 times faster and have 1000 times more endurance than NAND flash memory. Additionally, it could achieve storage densities up to 10 times higher than current memories. However, the first products to launch did not meet these expectations.

This new type of memory is constructed using phase change technology with a cross point that lacks transistors. Selectors and memory cells are positioned at the intersection of perpendicular interconnects, as depicted in the image above.

These cells, made of a material that has not been specified, can be accessed individually via a current sent through the top and bottom interconnects that touch each cell. To increase storage density, 3D XPoint cells can be stacked in three dimensions.

In the realm of the 3D XPoint storage technology, each cell has the capability to store one bit, meaning it can exist in one of two states, represented by a change in material property that modifies its resistance. The cell’s resistance can be either high or low, thereby representing binary data. As persistent cells, they maintain their values indefinitely, even when there is a loss of power supply, making them non-volatile.

Access operations such as reading and writing data are conducted by adjusting the amount of voltage directed to each selector. During write operations, a certain voltage is sent via interconnects to a cell and a selector, activating the selector and transmitting the voltage to the cell to initiate the general property change. In contrast, read operations require a distinct voltage to determine whether the cell has a high or low resistance state. This is how the memory functions.

On the other hand, it is important to note that 3D XPoint has a significant advantage over NAND technology. It can write data bit by bit, unlike NAND, where all bits in a block must be erased before new values can be written. In theory, this capability leads to higher performance and lower energy consumption for 3D XPoint than NAND flash memory.

Implementations

Both Micron and Intel have embarked on developing intriguing products using the new 3D XPoint memory technology. For instance in 2017, the first Intel Optane SSD DC P4800X, a 375 GB unit based on 3D XPoint, was released. The unit serves as an intermediate memory in systems equipped with Intel processors and compatible chipset motherboards. According to Intel, the P4800X unit performed five to eight times faster than the company’s NAND flash-based DC P3700, according to internal tests conducted on low-depth queues using a mixed workload. The P4800X can attain up to 500,000 IOPS, or around 2 GBps, with a queue depth of 11. In 2017, Micron QuantX, a storage memory based on 3D XPoint, was also launched.

Nevertheless, this 3D XPoint memory technology has not been widely accepted in the market due to several factors.

3D XPoint: advantages and disadvantages

Let’s explore some key features of 3D XPoint that should have led to its widespread adoption despite its limitations:

Unlike NAND flash memories, the 3D XPoint architecture does not require data to be stored in 4KB blocks. This file I/O stack is sluggish, but 3D XPoint streamlines everything by allowing for the quick and efficient access of small amounts of data.
Although 3D XPoint is not as fast as DRAM, it is quicker than NAND flash memory, which is a positive attribute, especially considering that it is also non-volatile.
It is a versatile memory that can use both M.2 formats to utilize a PCIe bus, as well as be integrated into DIMM modules.
However, despite these improvements, 3D XPoint’s expansion has been limited due to other drawbacks such as:
The new 3D XPoint technology is more expensive than traditional NAND flash memory.
The speeds projected by Intel and Micron have not been achieved. Some experts believe that this may be due to limitations in the PCIe bus used, which could be restricting its performance.
It also has other associated issues that require changes in the system to fully capitalize on its benefits. For instance, a compiler that can identify this memory as persistent is necessary so that applications can use the I/O system to perform accesses.

Cost

As previously mentioned, cost was a significant concern for 3D XPoint technology. For instance, the 375 GB Intel Optane P4800X was priced at $1520, which is equivalent to $4.05/GB. This is substantially higher than the cost of the 400 GB Intel SSD NVMe PCIe P3700 flash NAND-based unit, which cost $879, or $2.20/GB.

Therefore, when looking at the cost per gigabyte, 3D XPoint memory is almost twice as expensive as NAND flash memory.

Use cases for 3D XPoint

3D XPoint memory is employed as an intermediate memory or a type of cache between the main memory (DRAM) and secondary storage media (NAND flash such as SSDs) or HDDs.

Applications that are frequently accessed or that require high performance are stored in this intermediate 3D XPoint memory for quick access, while less frequently accessed or less performance-critical applications are stored in conventional secondary memory.

The pioneers of 3D XPoint memory envisioned extending its usage to more applications, but so far, this has not been the case. Even Intel has disposed of its Optane division, as you may know. Prior to this, Intel envisioned this memory being utilized for various purposes, such as:

Incorporating it into the virtual memory of an operating system to transfer processes that cannot fit into RAM, instead of storing them on slower drives in a pagefile.sys or SWAP.
Optimizing specific applications.
Data center servers and HPC with Intel Xeon.
Storing large databases to facilitate quick access.
Overcoming network bottlenecks in Big Data.
Facilitating high-performance computing applications.
Enhancing performance in cloud instances.
Using it as primary memory tiers in hyper-converged systems.