When Violin Memory, a pioneer in sold state storage arrays, made a Chapter 11 bankruptcy filing and announced an asset sale in December it marked another milestone in the company's slow decline. At its IPO in 2013 the company was valued at $162 million, but last year it was delisted by the New York Stock Exchange when its market capitalization fell below $15 million. Today it may be worth less than a tenth of that.
There are plenty of reasons why the company has not thrived in the way that it might have been expected to, and at least one of those is related to its strategy of eschewing commodity solid state drives (SSDs) in its high-end flash arrays, preferring instead to use a proprietary flash fabric architecture that utilizes a mesh of NAND flash dies organized into modules.
High performance solid state storage cards
On the face of it that's a good strategy for a company competing with the likes of Texas Memory Systems (now part of IBM) and Fusion-IO (which was acquired by SanDisk in 2014) for the top end of the market, as Jim Handy, solid state storage expert and semiconductor analyst at Objective Analysis, explains. "Custom modules get impressively higher performance for a slightly higher production cost," he says. "These custom cards provide much better latency and bandwidth than SSDs, and systems vendors can charge higher prices for that if they have the total solution that the client needs."
One of the problems for Violin appears to be that although it offers faster performance than SSD-based arrays using its higher-priced, proprietary, solid state storage modules, the extra performance isn't high enough to justify the extra cost. Other all-flash array makers offer better performance for those that really need speed, and cheaper, lower performance arrays that use standard SSDs suit most of the rest of the market.
Solid state storage services are key
But that isn't the only problem, according to Handy. "Systems vendors can charge higher prices (for better latency and bandwidth) if they have the total solution that the client needs," he says. "It’s really important to have the total solution in most cases. Violin isn’t very strong in services like replication, HA, et cetera, and this is a major drawback."
He contrasts this with IBM's competing offerings. "IBM has a big focus on services, and has been charging top dollar for systems that combine the low latency and high bandwidth of custom flash cards with services, high reliability, and a solid support network," he says.
Price rises for solid state storage
Even though the solid state storage market is expanding, that's not been a huge help to high end solid state storage array makers like Violin. That's because the solid state storage market is characterized by falling flash prices, making it affordable to run an increasing range of applications on solid state storage to get the benefit of increased performance.
But the indications are that the solid state storage market is going to change in character during 2017. That's because manufacturers including Samsung, Micron and Toshiba are making the transition from tradition 2D (planar) NAND to the more advanced and densely packed high capacity 3D NAND. 2D NAND production is expected to fall sharply as manufactures concentrate on ramping up 3D NAND production – and by Q3 2016 the share of 2D NAND is likely to be less than 50% of the global market.
The spanner in the works here is that yields for 3D NAND fabrication are not yet at the typical 80% - 90% level that you would expect for 2D NAND –particularly where Micron and Toshiba are concerned – and that means that the global production capacity for solid state storage is only likely to rise by about 6% in 2017, according to Sean Yang, a semiconductor research director at analyst DRAMeXchange, a division of solid state storage market intelligence company TrendForce.
But demand for SSDs is projected to soar by 60% in 2017, and SSDs account for almost half of global NAND usage, according to Yang. (Devices such as mobile phones account for much of the rest.)
And that can only lead to one thing, Yang believes. "Normally we see price declines in Q1 of each year, but we see SSD prices surging this year because of the shortage," he says.
Things are unlikely to improve in the foreseeable future because manufacturing 3D NAND has proved to be unexpectedly hard, he adds. "Only Samsung has made any progress – the other manufacturers are going to need some time before their yields increase – perhaps not until Q4 of 2017."
New production capacity takes a minimum of about two years to build, so there is no prospect of new fabrication plants being built in the near term to ease the supply shortage, says Yang. And longer term, as yields improve, supply will increase naturally without manufacturers having to make the considerable investment needed for new plants.
Delayed solid state storage adoption
What will be the result of this change in market conditions? If NAND prices go up then you would expect to see a drop in solid state storage adoption rates, because SSDs looks less favourable compared to cheaper spinning disk storage for applications that don't need a significant storage performance boost. "We are seeing that delayed adoption is happening already," says Yang.
Solid state storage vendors are also likely to find their margins squeezed. That's because the underlying storage economics hasn't changed, and storage buyers will only want to pay so much for solid state storage devices. (If vendors try to pass on NAND cost increases then buyers are likely to be unable to justify buying solid state storage devices. After all, applications that already require high performance storage probably already have it. It's falling costs that are enabling buyers to justify buying solid state storage for less performance-critical applications, and if prices rise then those less performance-critical applications can continue to run on spinning disk storage.
So who will be least affected by this margin squeeze? The answer is those storage vendors who have the greatest margins, and who therefore can absorb some NAND price rises while still turning a healthy profit.
And who might these vendors be? The ones that offer the higher performance solid state storage systems, especially the ones that use proprietary memory modules where the underlying NAND costs is a smaller fraction of the overall cost of the systems. Companies like Violin Memory, in other words. Now it may be too late for Violin Memory, but for the likes of IBM and SanDisk 2017 would seem to offer ideal trading conditions. "The focus will indeed have to shift as there won't be cost reductions for a while," says Jim Handy.
Newer solid state storage technologies
However he believes that increased NAND costs will also focus attention on standard, non-proprietary SSD technologies such as PCIe and NVMe that offer higher performance, allowing buyers to justify those higher costs. "PCIe has become as cheap as SATA, so systems will speed up simply by converting from SATA SSDs to PCIe/NVMe," he says. New technologies like 3D XPoint will also offer performance boosts.
In the medium term – 2018 and onward – NAND prices are likely to start to fall again, continuing the long term trend. When that happens then enterprise buyers will still be prepared to pay a premium for the very highest performance systems, but the majority of solid state storage buyers will be more interested in the price they pay for commodity SSDs – whether they are stuffed with 3D NAND or 3D XPoint. The question for the likes of Violin (or whoever buys its assets,) IBM and SanDisk will then be whether there is a long-term future in building solid state storage systems based on proprietary storage modules.