3D Xpoint Technology Development History

Micron and Intel began collaborating on the development in 2006. The Lehi, Utah facility was also built after Micron and Intel began working together in 2006. However, this new storage technology has had its share of industry ups and downs along the way. After a 10-year R&D cycle, Intel and Micron officially released 3D Xpoint technology in 2016.

In 2017, Intel launched the Aeon series of storage devices based on 3D Xpoint. In 2018, due to a disagreement on the development of next-generation technology, Micron and Intel reached an agreement to end their cooperation on 3D Xpoint after completing the development of the second generation of 3D Xpoint, each developing the third generation of products based on 3D Xpoint technology. 2019, Micron purchased Intel's stake in the Lehi facility, and Intel signed a foundry agreement with Micron to manufacture at the Lehi facility. But Intel's Aeon series was not sufficient to fully utilize the Lehi plant's capacity, and Micron had to suffer a loss of more than $400 million in non-GAAP operating profit each year as a result, leading Micron to eventually decide to sell the plant. However, Micron will still retain all of its 3D XPoint-related intellectual property on hand.

Looking back at the development of storage, 3D Xpoint is one of the most groundbreaking storage technologies since the introduction of NAND Flash. 3D Xpoint is seen as a disruptor in the storage industry due to four advantages.

• 1000 times faster than NAND Flash

• Half the cost of DRAM

• 1000 times the lifetime of NAND

• 10x the density of traditional storage

How 3D XPoint Works?

3D XPoint works on a fundamentally different principle than NAND. In contrast to NAND, which distinguishes between bits by pressing a certain number of electrons into an insulated floating gate, 3D XPoint is an impedance-based technology that distinguishes between 0s and 1s by changing the impedance of the cell through a large number of property changes. It consists of a selector and a memory cell, both located between a Wordline and a Bitline (hence the name Crosspoint).

3D XPoint

Loading a specific voltage value on the Wordline and Bitliane activates a selector that allows the memory cell to do a write operation (i.e., a large number of property changes in the cell media) or a read operation (allowing current to pass through, checking whether the impedance value of the memory cell represents high or low).

The industry is now generally pinning its hopes on EUV, and Intel and Micron are claiming that 3D XPoint will be compatible with EUV lithography, and that memory cell design sizes can be scaled down to a maximum of single-digit nanometer levels without significant impact on usage help/reliability. In fact, as physical size drops, it improves in some areas instead.

However, I'm afraid that volume production using EUVs will still not be possible for several years to come. The primary focus of the first EUV production will also be at the logic level, both because the cost of the equipment is too high and because logic cannot be vertically bloomed like memory, which can lead to heat dissipation issues.

In theory, 3D XPoint also supports multi-layer cell design, but Intel and Micron are not planning to pursue this route. While it is not too difficult to implement multiple layers of resistance in the lab, it is far more difficult to ensure that each die has the necessary characteristics to operate as a dual-layer cell in the tens of thousands of wafers produced.

Knowing the basics of how 3D XPoint works, it may seem simple. But the reality is far more complex than described above. In particular, there are special voltage differences and specific materials, the principles of which no third vendor has yet mastered.

Applications of 3D XPoint Technology

Thanks to the advantages of 3D Xpoint technology, it can be used in a wide range of applications such as gaming, media production, genome sequencing, financial services trading, and individualized therapy. The following are some examples of 3D Xpoint applications that show the potential for future applications of 3D Xpoint.

The most obvious application of 3D XPoint will be as an intermediate layer between DRAM and SSD: Throughout the history of computing, many intermediate layers such as on-chip cache, off-chip cache, cache SSD, etc., have been built between memory and processors. 3D XPoint memory as a storage medium fits well into this layered structure and can fill the gap between DRAM and today's fastest non-volatile storage. When used as another layer of cache, 3D XPoint can further accelerate applications that are currently limited by capacity or memory latency.

Servers using 3D XPoint: Intel and Micron can't wait to point out the technology's use in "big science" projects and systems such as the Large Hadron Collider and Oak Ridge's Titan supercomputer, which generate huge amounts of data. However, the processors are in trouble, and how to handle all this data is the first and most important requirement, as is getting the data to the processor. Any analytics system that benefits from this mechanism could benefit from 3D XPoint if standalone processors could access data pools of SSD-level capacity as well as DRAM.

The financial industry is likely to be the first to adopt 3D XPoint: Because the financial industry is most inclined to use significant technology to put itself ahead of the curve in a highly competitive and lucrative field. From this perspective, 3D XPoint will not add much processing speed, as data processing is already deployed into large-capacity DRAM pools whenever possible, but 3D XPoint will allow traders and analysts to run large data set simulation programs more efficiently.

Since 3D XPoint is an unprecedented technology, it is impossible to think of all potential future applications for 3D XPoint at once. 3D XPoint has the potential to change modern computer architecture and the way we look at computing, but this transition will not happen overnight and may also require other vendors to have the appropriate competing technologies to meet the demand. What is clear, however, is that 3D XPoint is likely to take us into a new era of memory and computing.

While DRAM and NAND have improved and shrunk a lot over the decades, the release of 3D XPoint still counts as the biggest event in the storage industry not seen since the invention of NAND in 1989. 3D XPoint is actually a whole new kind of memory. Fast, durable, scalable, and non-volatile, DRAM and NAND can only meet two of these standards each. 3D XPoint fills the gap between DRAM and NAND, leveraging the best features of both technologies to create a type of memory we have never seen before.

Technical challenges still to be overcome with 3D XPoint

Micron's decision to withdraw from 3D Xpoint has a significant relationship to the poor commercialization of the technology. It is known that only Intel's Aeon series currently uses 3D Xpoint technology. Micron has announced several storage devices based on 3D XPoint flash chips, such as the X100, but they have never been officially launched. In addition, 3D Xpoint has not been successful in establishing a sound industrial ecosystem. Micron's chief commercial manager Sumit Sadana said in a media interview that many customers were not satisfied with the X100 because they had to rewrite most of the software to take advantage of the new memory, so the product was offered in very limited quantities.

In the future, 3D Xpoint technology will be how the development of considerable attention. It is reported that Intel will produce 3D XPoint chips for its own Aoton product line in its fab in New Mexico. Although Intel has sold its NAND flash memory business to SK Hynix, it may still take over Micron's Lehi factory to ensure the supply of its own product line. After all, Intel still offers products related to the use of 3D XPoint technology in enterprise product lines such as data centers. However, if 3D Xpoint only stays at this level of commercialization, the future of this new storage technology is still not clear enough.

At this point, there are still some challenges to overcome if 3D XPoint is to be mass-produced. First, 3D XPoint requires the use of approximately 100 new manufacturing materials. Some of these raw materials are currently in very limited supply, so the supply chain needs to be carefully adjusted.

Second, because 3D XPoint requires more processes, processors will need to increase their plant space and initial capital by a factor of 3 to 5. For example, the production of the first generation of 3D XPoint memory requires 2.5 square meters of wet processing equipment capable of processing 180 wafers per hour, while the second generation of 3D XPoint requires the same equipment to process 1,000 wafers per hour, 3D XPoint increases the processing plant for plant and capital requirements. In addition, 3D XPoint also requires more productive equipment.

All of the above challenges increase the cost of 3D XPoint. However, for the 3D XPoint market, the cost is key. Second-generation 3D XPoint allows for a four-layer cascade at approximately half the price of DRAM. The current market trend is that if the cost of 3D XPoint is not half that of DRAM, consumers will prefer DRAM to 3D XPoint. Since DRAM is still decreasing at a rate of about 30% per year, it will be difficult to keep the cost of 3D XPoint half that of DRAM.

In addition to 3D XPoint's own technical difficulties, the challenges from other new memory technologies should not be underestimated.

For example, software NVDIMM-P is a strong challenger. Software NVDIMM-P uses software algorithms to predict the frequency of data access and store the data in DRAM (for high access frequency data) or NAND (for low access frequency data) based on the access frequency prediction. Such a technique balances cost and performance, as DRAM is faster but costly to access, while NAND is less costly and denser but slower to access. Obviously, the overall performance of software NVDIMM-P depends on the software algorithm and the application, in some applications the prediction of software NVDIMM-P algorithm is more accurate but in other applications, the access frequency of data is difficult to predict.

3D Super-NOR is also very promising. 3D Super-NOR technology uses 3D stacking technology and offers very low latency. BeSang, the manufacturer of 3D Super-NOR, claims that 3D Super-Nor can be achieved at a cost 10 times lower than 3D XPoint, but we still need to wait for 3D Super-NOR to be produced in volume to see if it can replace 3D XPoint. 3D XPoint.

The market potential for 3D XPoint is huge, but its complex manufacturing process results in high costs, making it difficult for 3D XPoint to replace DRAM in the near future, but we should remain optimistic.