In terms of up-and-coming investment grade technologies, flash memory seems like it ought to reside low on the list.
First off, it’s about as new as the Ronald Reagan presidency, having been initially developed by a Toshiba engineer in 1980 as a way to maintain stored data after users turn off a computer. A few presidents later, flash technology is ubiquitous in iPods, digital cameras and other devices, and it boasts a supply chain dominated by massive electronics conglomerates.
As for profit margins, flash prices are currently declining at a rate not too different from the sort of commodity play venture investors usually shun.
But in looking at recent funding activity, it’s clear that flash—along with a cadre of next-generation memory technologies—is generating broad appeal with venture investors. Since last summer, VCs have invested more than $230 million in memory-related deals, according to Thomson Reuters (publisher of VCJ).
The funding mix includes some large rounds. Fusion-io, a Salt Lake City-based maker of enterprise storage devices, announced in April that it raised $45 million from Accel Partners, Andreessen Horowitz, Lightspeed Venture Partners, Meritech Capital Partners, New Enterprise Associates and Triangle Peak Partners.
Flash is appealing to VCs because prices have come down enough for the technology to extend beyond consumer devices and into the enormous enterprise and data center markets, says Chris Schaepe, managing director at Lightspeed. More broadly, memory investments are driven by the momentum of quickly escalating demand.
“The bottom line is there’s way too much data, and it’s pushing the limits of traditional analytics and warehousing,” says Larry Orr, general partner at Trinity Ventures, which led a $20 million financing round in October for GridIron Systems, a Sunnyvale, Calif.-based maker of appliances to provide faster access to information in data centers. That trend of increasing data filling up storage warehouses, he says, is motivating investors to scope out startups with credible strategies to alleviate the crunch.
VCs aren’t forgetting the massive scale of the memory business either.
The bottom line is there’s way too much data, and it’s pushing the limits of traditional analytics and warehousing.”
Global memory revenue will likely reach $57 billion this year, a 28% rise over 2009, according to research firm Databeans, which reports that the sector accounts for 22% of the $259 billion semiconductor market. The firm predicts memory revenue will grow an average of 12% annually until it reaches $99 billion in 2015.
Currently, venture-backed companies have a limited role in that ecosystem, as virtually all product sales go to mature manufacturing companies, such as Samsung and Toshiba, the two biggest flash producers. However, startups are finding room to make a mark in other ways, typically by developing technology that enables end-users to deploy memory more effectively and cheaply.
For example, in data centers, many server processors are underutilized due to bottlenecks between main memory and disk, according to consulting firm Web Feet Research. To overcome this issue, data center operators and their customers are seeking faster storage and more efficient memory architectures.
That’s where Fusion-io comes in. The company, probably best known for having Apple co-founder Steve Wozniak as its chief scientist, develops flash-based solid-state memory technology that it markets for use in server systems as a means to more efficiently draw data from storage infrastructure. In a case study published on the Fusion-io website, the company claims its technology enabled Answers.com to run more than 9 times as many queries per second.
Investors like what they hear. To date, the company has raised $111.5 million from VCs, including new investor Meritech Capital Partners, which led the latest round. Fusion-IO is also already generating significant revenue, with quarterly sales figures topping eight digits, according to a March press release and 80% quarter-over-quarter sales growth.
For other flash companies, rising prices of competing memory technologies are pushing adoption. Research firm iSuppli predicts that with the economy heating up, global revenue for dynamic random access memory (DRAM) will grow to $31.9 billion this year, up 40.4% from $22.7 billion in 2009. Part of that is due to rising prices, putting a squeeze on PC makers who are its primary consumers.
Schaepe of Lightspeed says that pricing trends will enable flash to gain significant share in enterprise applications, as it is cheaper than DRAM but has similar performance characteristics.
While flash prices are declining, overall sales are rising. The flash market exceeded $25 billion last year for NAND and NOR, the two primary types of flash memory, according to Objective Analysis, a Los Gatos, Calif.-based research firm. NAND flash, commonly used for storage in devices such as USB memory sticks and digital cameras, accounted for about $21 billion. Spending on NOR flash, often part of cell phones, totaled $4.5 billion.
The question for us is: After flash, then what?”
Objective Analysis predicts NAND will grow to about $31 billion this year and that NOR will grow to about $5 billion. While enterprise purchases of flash-based solid state drives (SSDs) currently make up only a small portion of flash expenditures, they’re growing at a rapid rate. One subsector—enterprise for flash-based SSDs—should reach $510 million this year, and will grow at a 55% average annual rate to reach $3.8 billion by 2015, Objective Analysis predicts.
Ensuring a smooth fit with legacy systems remains challenging.
“If you introduce flash, it’s like giving a race car driver an airplane,” says Schaepe, one of several partners at Lightspeed who have carved out a niche looking for technologies to make flash of suitable quality and with the performance characteristics needed for the enterprise.
Enterprise flash drive developer Pliant Technology is one of the players in that space. The Milpitas, Calif.-based company, which has raised about $55 million since 2007 from Lightspeed, Arcturus Capital, Divergent Ventures and Menlo Ventures, makes solid state data storage drives that it markets to data center customers to add to their existing storage and server systems. While traditional SSDs and disk drives can only perform one read or write at a time, Pliant maintains that its drives can read and write simultaneously, enabling faster performance.
Yet convincing large enterprises to adopt new suppliers and technologies is often an uphill battle in the memory space, says Orr of Trinity.
“The irony is that unlike most cases of economies of scale, there is almost a ‘dis-economy’ of scale for memory—in the sense that data centers have huge amounts of memory that are using systems like EMC and IBM’s,” Orr says. Commonly, cost of storage for enterprise is actually much higher than for “the commodity stuff you and I have on our disk drives,” he says. That’s in part because what customers are paying for is not so much the storage as for added features, such as backup and security.
There are exceptions, Orr notes, such as Google, which has invested heavily to develop a scalable infrastructure of commodity servers and storage. “But 99% of companies are not in a position where they can invent their own proprietary architecture,” he says. “That’s why I think there’s a lot of opportunity.”
Trinity’s most prominent memory investment is in GridIron Systems, which raised $20 million in December and counts Mohr Davidow Ventures and Foundation Capital among its backers. The draw for Trinity, Orr says, is that GridIron is not attempting to replace storage subsystems of incumbent vendors. Rather, its appliance can offer performance enhancements while working within the existing infrastructure.
If you introduce flash, it’s like giving a race car driver an airplane.”
In the case of flash, the other obstacle to adoption has been price, says Kent Langley, CTO of data center operator nScaled. “To date, the prices of flash-based devices have been too high,” Langley says. “But I do expect to see our use of flash-based storage to increase over time as the devices continue to become more affordable.”
In the past few years, venture investors haven’t logged blockbuster exits for investments in enterprise flash and next-generation memory. Yet that could change as top-performing startups reach the kinds of revenue milestones typically required for a public offering, says Barry Eggers, a managing director at Lightspeed.
Appliance plays also have a history of warm reception in public markets, says Orr, who compares GridIron’s model to that of storage system provider NetApp, which went public in the mid-1990s and now has a market cap of about $11.7 billion. “At the time it came up, it was a standalone performance appliance, and the company has performed well after IPO,” he says.
Yet no one’s expecting memory investments to mimic the hyper-growth trajectories of many successful Internet startups. Particularly for early stage investors, these are projects expected to take at least several years to exit.
Long-term horizons are especially prevalent among investors in next-generation memory technologies, which they hope will exceed flash in many performance capabilities.
“The question for us is: After flash, then what?” says Andy Rappaport, a partner at August Capital and an investor in Unity Semiconductor, a developer of storage-class memory products employing a low-cost, non-volatile memory technology it developed called CMOx. The Sunnyvale, Calif.-based company raised a $22 million round last year from August, Lightspeed and Morgenthaler Ventures. Rappaport estimates the company has raised close to $60 million to date.
Much of his interest in the investment, Rappaport says, revolves around the notion that “while flash does a lot of things well, it doesn’t scale.” At some point soon, he predicts, prices will stop dropping at the steady rate required to continue seeing broader flash adoption.
Another up-and-coming memory technology that’s attracting VC is magnetoresistive random access memory (MRAM), a method of storing data using magnetic charges, rather than electrical ones. Though the technology dates back to the 1990s, startups have gravitated to the technology more recently, including two—Everspin Technologies and Avalanche Technology—which raised rounds in the past two quarters.
Fremont, Calif.-based Avalanche, founded in 2006, raised $7.5 million in December from Bessemer Venture Partners and Sequoia Capital, while Chandler, Ariz.-based Everspin, which spun out of Freescale Semiconductor, has raised $42.5 million since 2008, including $10 million in February from Draper Fisher Jurvetson, Lux Capital, New Venture Partners and Sigma Partners.
To date, a key source of MRAM’s appeal is that it doesn’t wear out or require power, says Stephen Socolof, managing partner at New Venture Partners. Socolof says the technology is currently used in casino gaming devices, power meters and “other applications out in the field that you just want to run indefinitely.”
As for pricing, “MRAM is still early on the curve that flash has gone down in terms of density and cost,” Socolof says. “The question is: Can it become dense enough and cheap enough to become universal memory?”