Toshiba, SanDisk Piloting 3D NAND That Doubles Previous Capacity

Lucas123 writes: Under a joint development agreement, Toshiba and SanDisk have begun pilot production of a new 48-layer 256Gb NAND flash chip in a brand new fab in Mie prefecture, Japan. The new X3 chips, which double capacity from 16GB to 32GB over the previous product, are made with triple-level cell (TLC) flash compared with Toshiba's last multi-level cell (MLC) chip, which stored two-bits per transistor. The chips are expected to begin shipping in products next year. The companies plan to use the new memory in a wide number of products, including consumer SSDs, smartphones, tablets, memory cards, and enterprise SSDs for data centers, the companies said.

Comments

Will not buy TLC NAND

[D.McG.]

The data longevity of 3-bits per cell NAND is quite poor. MLC's 2-bits per cell still has me uneasy. Storing 1 of 8 voltage levels in an environment that leaks electrons over time is not for me. I'll wait for XPoint before upgrading my MLC SSD.

Re:Will not buy TLC NAND

[serviscope_minor]

You know I thought you were wrong and that a Triple Level Cell had 3 levels per cell, storing about 1.6 bits. Apparently I was wrong. TLC despite standing for triple level cell apparently means in fact 8 levels per cell where as MLC (multi-level) is for 4 levels per cell.

Whoever came up with these terms is nuts.

Re:Will not buy TLC NAND

[Rockoon]

Makes perfect sense to me. Maybe its because I understand binary.

Re:Will not buy TLC NAND

[amorsen]

Did you think SLC (single level cells) had 1 levels per cell, thereby storing 0 bits?

I own one of those. It reads zero in all locations.

Re:Will not buy TLC NAND

[Kjella]

Well, smaller process sizes also reduce write cycles. When they went from 2D to 3D they went back to a bigger process size with less defects due to all the layers. Though they had some controller/firmware issues the first 3D TLC NAND had more raw write cycles than state of the art planar MLC NAND. Of course now they're shrinking it again in the quest for even more storage, but the clock got a pretty good reset going from 1 layer to 48. Going from MLC to TLC is more of a variation that cuts write cycles to about 1/3rd for 50% more space, all other things being equal.

Re:Will not buy TLC NAND

[serviscope_minor]

You had posted that you just discovered that it is, and now you are posting that it isn't.

Try actually reading what I wrote before being overcome with smugness. MLC can measure multiple (i.e. > 2) voltage levels on the floating gate transistor. The number of bits is the log_2 of the number of distinct voltage levels.

Calling something with 3 BITS that has 8 LEVELS triple level is silly. Why not just call it 3 bit cell. SLC sort of made sense from an EE point of view as being able to store one level as distinct from ground, but 1 level distinct from ground (1 bit) does not make 3 levels distinct from ground 3 bits.

Back in my day

[Scottingham]

Our storage mediums spun and made noises! And we liked it!

Re:Back in my day

[marciot]

Our storage mediums spun and made noises! And we liked it!

What is funny is that I recently upgraded my laptop to an SSD and I was flabbergasted that during disk access it made the exact same noise as it did when I had a regular hard disk. I thought I had been sold a fake SSD, but I’m getting approx 510MB/s on benchmarks, so I know it is legit. On further investigation the noise comes from where the speakers are located, so what I thought was hard disk noise was probably just electrical noise. My system makes the same soft buzzing sound on data access as it always had!

Re:Too bad

[BenJeremy]

XPoint won't be 10x the price - more like 3x~4x the price of consumer flash when it hits the market.

The economy of scales will drive production up as it is incorporated into mobile devices and enterprise systems, which will also drive down price. XPoint is really more a matter of dooming platter drives to extinction, because the durability, power consumption, and speed will make it highly desirable in the enterprise market.

I wouldn't be surprised if XPoint hits the "tipping point" of $200 for 512GB within 2 years of production, making it a prime choice for enthusiasts. Meanwhile, SSDs will continue to edge out platter as the "cheap" mass storage option for most consumers.

We've seen Samsung 512GB drives hit $135 shipped to consumers recently. On Black Friday? Maybe we'll see them hit $100 or less. Once this 3D NAND hits full production, expect those prices to plummet. Meanwhile, platter drive makers are still trying to eke out every dollar with artificially propped up pricing, caught in a bit of a quandary... consumers rarely need more than 2TB, but the lowest you can produce a drive... any drive... is around $30; $/GB is not as scalable for platter drives as it is for SSDs (or XPoint), because you have a minimum BOM to fill before you can get a platter spinning and reading data, a baseline cost. Higher capacity drives can make you money, but consumers aren't buying them. Yes... you can sell $40 2TB drives all day long, but stockholders don't like that sort of crappy profit margin, so they continue propping up the price on a 5 year old disaster (Thai floods) as SSDs quickly play catch up on pricing with a far more compelling product. [Insert buggy whip story here]

Re:Too bad

[mlts]

What HDD makers really need to do is stop focusing on price and make a line of drives that is made to be archival grade. For example, there was a line of drives with two read/write heads that worked in an active/active configuration.

What might be even better would be to make a standardized, rugged drive cartridge case, similar to iMation's RDX. Something that can handle drops, be gripped easily by a tape silo's robotic mechanism, can handle tens of thousands of mounts/dismounts, has built in encryption, the ability to have WORM functionality (similar to late gen DLT drives where the cartridge can be formatted as normal or WORM), and so on. The drive can be presented either as a tape volume, standalone JBOD hard disk, or part of a RAID set (and inserted/ejected at the same time with 2-3 companions.)

Moving HDD to a backup/archive use as opposed to primary storage will keep this technology relevant, as opposed to trying to fight with SSD (which is a better primary storage technology [1].)

[1]: In all ways but recovery. An SSD goes bad, there is no way to recover the data, period.