See also: New memory technology to replace NAND
Unity Semiconductor, the US start-up with a non-volatile memory technology scalable to 20nm, is looking for a partner to bring the technology to market.
The seven year-old company, which emerged from stealth mode earlier this week, is giving almost as much thought as to how to bring the technology to market profitably, as it did to bringing the technology to the point of commercialisation.
One of the worst problems about bringing a new memory technology to market is the fear that it will become commoditised and unprofitable.
If you go down the IP licensing route you're up against loads of potential trouble a la Rambus. If you compete head-for-head as a device supplier you're up against an over-supplied, chronically unprofitable, market.
"If you look at the current business model for memory it's unsustainable because it's not generating cash," Darell Rinerson, founder, chairman, president and CEO of Unity, told Electronics Weekly.
Unity has two strategies for avoiding commoditisation:
One way is the traditional method of extensive patent protection - Unity has 60 issued and 90 pending patents which will be combined with a 'selective licensing' policy which will control the spread of the technology.
"We believe we have an IP position to keep non-licensees out of the business", said Rinerson, "the IP will be quite concentrated. The ability to keep others out and control the business is achievable, we don't want it over-commoditised."
The other way by which Unity hopes to avoid its technology becoming commoditised is through an innovative approach to manufacturing.
Unity's wafers will be made in two different fabs. The first three layers of metal on a wafer will be made on a CMOS logic process at a Japanese fab. Then the memory layers will be added at a fab with a 35nm process which is, as yet, not chosen.
Unity's manufacturing model involves the separation of the processing of what it calls the Front-end-of-line (FEOL) CMOS base wafer from the Back-end-of-line (BEOL) memory layer processing'.
"No new process technology is needed in the CMOS base wafer, which can be fabricated at a CMOS logic foundry with existing production capability and capacity on a trailing edge CMOS (90 nm) process," said Rinerson.
"The BEOL memory concept enables a CMOS logic foundry to be in the memory business without taking significant risks associated with being in the memory business."
The strategy allows Unity to be a moderate follower in CMOS transistor technology. Its shrink path is unconventional in that a higher density memory core doesn't require base CMOS technology migration.
Instead, Unity can use the same 90nm base CMOS for multiple generations, as well as use proven design IP to reduce risk and time to market.
"Logic vendors would love to be in the memory business but haven't been able to do that because memory processes are very different to logic processes", said Rinerson.
"We have entered into a relationship with a Japanese company with its own 90nm CMOS logic process to make the base layers (three layers of metal) for the 64G memory", said Rinerson.
The idea is that the wafer then goes to another company's fab (as yet not chosen) which has (or will have) a 35nm process to put on the memory layers (4 physical layers).
The idea of splitting the front end processing into two stages is to try and prevent commoditisation of the technology with no one fab having the capability to make the whole chip
Unity is currently open to offers for partnership for the fab which will do the memory layers.
"Samsung is a potential candidate. All the leading memory manufacturers are involved", said Rinerson, "Micron is a possibility to partner with."
Why not IBM? "The IBM business model is not going into high volumes," replied Rinerson, "but it's doing foundry work which is quite lucrative. We're looking for companies willing to work with us on bringing it into production."
There is a possibility that the right kind of deal with a partner might help bring in, or defray, some of the $25 million it will cost to get Unity through to Q2 2011 when it is set to produce its first commercial silicon.
Unity is different from most start-ups in that it never had a Eureka moment. It didn't stumble across a scientific effect and think 'Wow! This could be the Next Big Thing'.
Instead, Unity set out to look for a technology which could out-perform the mainstream chip storage technology which is NAND.
The big motivation was that NAND, based on floating gate technology, is seen as running out of scalability beyond 32nm.
"In order to beat NAND we went out and looked for a technology which could implement a cross-point switch", said Rinerson, "we were very intrigued by the concept of oxides. We thought the effect of oxides would be simple to understand."
Now he knows it's not. It's taken seven years and some $50 million to bring the technology to the point where it can be commercialised.
Rinerson originally found the technology in work being done at IBM's Zurich R&D Centre and the University of Houston.
Seven years later, Unity has engineered a passive rewritable crosspoint memory array that requires no transistors in the memory cell
"We are the only company in the world that can write and read data patterns in a passive cross-point array," says Rinerson.
Why has it taken so long? "Things always take longer than you think," replied Rinerson, "the switching effects were observed, but the reason for these effects were not understood. You have to understand the physical concepts. What we've been doing is to develop a good understanding of the physics behind the switching effects."
"The switching effect occurs in layers of oxides which are not standard materials used in the semiconductor industry", continued Rinerson, "it is based on the movement of ions, specifically oxygen ions." Drily, he adds: "It is somewhat difficult to engineer the effects".
Will it scale? "Anyone working on a technology is always optimistic", said Rinerson, "scaling is something which you have to do to find out what the problems are. We feel, we believe, with reasonable confidence, this will be scalable. If Unity has truly identified the memory effects, then CMOx (Unity's name for its technology) is the best candidate to continuing memory scaling."
Now the race is on to get a product to market. Unity says it will tape out a 64Gbit memory on a 35nm process in Q2 next year. It says it will be in production in two years time.
Unity says CMOx will scale to 20nm and delivers a density 4X that of NAND and five to ten times faster.
2012-13 is scheduled for 128Gbit and 256Gbit memories, also made on 35nm.
Then, with the transition to 20nm in 2015, comes the milestone of a 1Terabit memory IC.
Rinerson, formerly of Micron and AMD, has raised $75 million. The company recently closed its Series C round of financing of $22 million.
The financing came primarily from August Capital, Lightspeed Venture Partners, Morgenthaler Ventures and a hard disk drive (HDD) manufacturer. That brings total funding to nearly $75m. The company needs another $25 million to get the technology into production and some could come from its production partner or licensees.
The CMOx manufacturing process produces a wafer which is more expensive than NAND, but the density achievable with CMOx is claimed to be four times the density achievable with NAND.
"The mask count will be only modestly higher than the mask count for NAND," says Rinerson, "and the wafer cost will be 1.5X more expensive than NAND, but the die cost will be less than NAND because the density is 4X NAND density for the same area of silicon."
The endurance of CMOx is the same as the endurance of NAND, says Unity. with NAND quotes 10,000 to 100,000 cycles. So CMOx used in SSDs will have to use the same wear-levelling algorithms as NAND.
Unity does not intend to make SSDs or modules. The intention is to be purely a chip company.
Why has Unity emerged from stealth mode now? "We think that the two year horizon to production is a good time to come out of stealth mode", says Rinerson, "we haven't published technical papers. We didn't want to attract competition for IP. Now we feel our IP position is very strong."
See also: Mannerisms, the blog of David Manners. Updated twice daily, it's the distinctive, entertaining, authoritative and never dull commentary on the semiconductor industry, from someone who knows. Sign up for the Mannerisms eNewsletter.