3D xpoint, kioxia says this technology 'is not the future'

Kioxia claims that Storage-Class Memory (SCM) like 3D XPoint is not the future and has no long-term prospects. This was exposed by the company at the recent IEDM conference, where the company promoted its BiCS flash and XL-Flash technologies. The reason would be the low cost per scaling bit as the number of layers increases.

Kioxia says 3D XPoint technology 'is not the future'

In the last half of the past decade, the flash industry switched to 3D NAND technology. Compared to conventional NAND, the cell size was reduced, but this was more than compensated by stacking the layers vertically, for example, with 96 layers in the latest 3D NAND from Kioxia.

Also in the second half of the past decade, Intel and Micron started talking and shipping SSDs and DIMMs Optane. Optane is based on what companies call 3D XPoint, a form of phase change memory (PCM). 3D XPoint also uses stacking (hence the name 3D), but the first generation is only two-layer. The next second generation of 3D XPoint will double this amount to four layers.

A big difference and advantage of 3D NAND is that many layers can be processed simultaneously, so the production cycle time of a chip or 3D NAND wafer would not be as long. However, this is exactly the case in 3D XPoint, where each layer has to be processed individually. Although this does not duplicate the cost of the entire chip at all (or quadruple in the case of four layers), it means that the prospects for scaling from 3D XPoint to a large number of layers are limited.

This is the argument used by Kioxia to criticize 3D XPoint. The graph shows the cost in bits of the number of layers function of 3D SCM or NAND flash memory, normalized compared to one layer of memory. Unsurprisingly, for 3D NAND the benefits of scaling are impressive, producing a ~ 10x reduction in cost per bit for 16 layers.

Visit our guide on the best SSD drives on the market

On the other hand, storage class memory doesn't perform very well, and it even starts to increase beyond four layers, and reaches cost-per-bit parity with a single layer to 14 layers. At its most efficient four-layer point, it produces a cost reduction of 40% of the bits, far superior to the cost reduction of ~ 70% of 3D NAND in 4 layers.

Of course, this is data provided by Kioxia that, at some point, sweeps a bit for home, but that gives an idea of ​​why the implementation of 3D XPoint memory is not being so popular. We will keep you informed.

Tomshardware font