26-10-2020 | | By Sam Brown
Recently, Intel has announced that it will sell its NAND business to SK Hynix for $9 billion. Why is Intel selling off its NAND business, what is Optane technology, and what does the future of solid-state memory look like?
Recently, Intel announced that it would sell its NAND business to SK Hynix, a South Korean company specialising in memory technologies for $9 billion. While the deal still needs to be approved by regulators, the agreement will see SK Hynix have access to Intel’s NAND SSD technology, employees, IP relating to SSDs, and the NAND fabrication facility in Dalian, China for an initial sum of $7 billion. The sale of IP relating to wafer construction, R&D employees, and factory workforce will be transferred after a $2 billion payment in 2025.
During this period, Intel will retain the wafer production capabilities of the foundry in Dalian, meaning that SK Hynix won’t be receiving the full business until 2025. As a result of the announcement, SK Hynix stock has fallen by 1.73% as investors feel the deal is in Intel’s favour (remember that SK Hynix will have paid $7 billion without full control until four years later).
Intel has had some major drawbacks in the past few years relating to a decision made by Intel decades ago; linking the node technology with the latest CPU architecture. To put it simply, the node technology, which relates to the smallest feature size producible on a semiconductor, would only be released when Intel released updated versions of its architecture. This means that Intel’s architecture releases were tied to its ability to shrink transistors, and this was not an issue for designs above 100nm. However, Intel has hit several technological challenges with their most recent shrinkage to 7nm, causing delays in new CPU designs. The result has been that their major competitor, AMD, has taken over who now produce the most advanced CPUs.
While a business can have multiple fields of speciality, focusing on too many different industries simultaneously can make it hard for specialisation in any one particular field. While memory was in fact one of Intel’s first products, they are mostly known for their processors. This has lead to Intel specialising in wafer fabrication, processor design, and development of memory technologies including Optane, SSDs, and NAND flash. According to Intel, the NAND memory business sector is being old off as it’s NAND memory business is underperforming and holding Intel back. However, the deal will not be selling off Intel’s Optane business as this was mostly being sold to a different company, Micron, in 2019.
Optane is the code name given by Intel to their 3D memory technology that uses PRAM, or Phase Change Random Access Memory. PRAM has the benefits of RAM being much faster than FLASH with lower latency while also being non-volatile, thus retaining its data even when the power is removed.
Unlike traditional memory systems, PRAM utilises a special glass-like material called Chalcogenide glass, which is a special glass that uses sulphur, selenium, and tellurium but lacks oxygen. By using electricity, the glass structure can be transformed between an Amorphous crystal and a polycrystal structure, and this changes its electrical resistance. While PRAM does have a maximum number of write cycles, it is generally around 100 million (as compared to 100,000 for typical FLASH memory cells), and thus can be used for greater amounts of time. The technology pioneered by Intel, called 3D XPoint, takes PRAM technology and creates a 3D structure which not only helps to increase memory density but also allows for the creation of high-speed, low latency SSDs that rival those of FLASH.
Currently, FLASH memory plays an important role in SSD technology, which allows for systems to gain a significant speed boost when compared to standard mechanical HDDs. However, even with the inclusion of FLASH, the number of stages between stored data and the processor introduces layers of latency that essentially prevents a system from operating at its maximum potential. These layers, however, are introduced due to the advantages of memory at each stage. For example, magnetic storage memory (such as hard drives), provide incredibly large amounts of memory at a meager price and are ideal for storing data long term. Cache memory, which stores data between the CPU and RAM, is incredibly fast and allows for the CPU to operate on information at full speed, but the cache is small as it is expensive to produce.
However, one memory type that researchers have been hunting for is what is known as “universal memory”. Universal memory would essentially combine all memory in a system into one storage location. Such a memory would be non-volatile, quick, and low cost so that it can allow for maximised CPU speed while storing large amounts of data indefinitely. Unlike FLASH, it would not degrade with use, and a system would not need a cache, RAM, or external storage. PRAM could provide future designs with such a universal memory. However, its maximum write cycles combined with its lower memory density than FLASH means that it is still not at the universal memory level.