Sep 18

The Way Forward

Developing any new technology is expensive, revolutionary technologies even more so. It is expensive to introduce new methods, materials and production techniques etc.. It is not enough just to perfect all these but one has to show the markets a compelling reason (price, performance) why your product (or solution, as current marketers like to say) is superior to the technology you are attempting to replace. So, by definition, revenue from a new technology is essentially zero until you have spent an awful lot of money. The memory business is no different.

I can point to three ways that companies achieve these goals.

intel micron1) Be a big company with an extensive product portfolio and cash flow. The development of 3D NAND is a good example. More recently Intel/Micron 3D XPoint, whatever the technology actually is, is an example of a radical departure from existing non-volatile (now called persistent, for some reason) memory technology. Intel/Micron, whether by design or necessity, have fudged the issue of incumbent technology replacement by stating that 3D XPoint actually fits in between two incumbents, DRAM and NAND. A risky strategy which I’m sure Intel/Micron competitors will not leave unaddressed.

2) If you are a small company you, by definition, don’t have the resources (cash) to pursue 1). So you can generate the necessary resources by using investment to fund the development costs of your ‘secret sauce’. This can be seen as a traditional start-up route where you perhaps start with some government funding at, for example, an academic institution, form a start-up with some Angel funding and then go the Venture Capital route.

crossbarchallengeCrossbar is a good example of this approach as are many other start-ups. They spring to mind as they have recently announced a Series D (I believe this means 4th) round of VC funding to the tune of $35M. All credit to them. Nobody gives money away these days and VC’s are no exception. I’m sure some very difficult negotiations and a large amount of sleep deprivation went into closing this deal. The question that springs to mind is, not so much, ‘Is the money raised enough to develop their vision of a 1TB stand alone memory chip?’ (It is not, I can ensure you) but rather ‘What is the specific level of technology maturity that can be achieved/has been promised for the $35M?’ Of course, I can only speculate as to the answer to this second question but presumably the maturity should be such that it is attractive to one of the volume memory manufacturers.

3) A variant on 2) is to use your investment to buy yourself a cash flow and potential market for your new technology. Adesto Technologies is such an example. Some time ago, they purchased the flash memory side of Atmel which provides NOR type low density memory across a large application space. They are in the news as they have made the necessary filings to pursue an IPO, i.e. become a public traded company. They hope to raise $50M based on their current business which is derived from the products they have developed based on Atmel’s technology.

maverickCBRAM, now called Mavriq (as an aside, I think this naming is just terrible, who on earth will want to risk using a maverick technology in their mass produced device?) is touted as “ideally suited for applications that require high-performance while consuming 1/10th to 1/100th the energy of existing flash memory products” with no specific revenues associated with it. Although Adesto are clearly currently a loss making concern, their reason for optimism for success of the IPO is based around future sales based on “The number of our design wins has grown from 32 in 2013 to 65 in 2014 and to 88 in the six months ended June 30, 2015.” Not given is the number of Mavriq design wins.

Good luck to All!

Christie Marrian, ReRAM Forum, Director
ps Ford Maverick, geddit?


  1. Carlos Araujo

    Your 1-4 analysis is useful but is lacking two important components: (a) True breakthrough in ReRAMs has yet to be realized. So, it is still possible to develop IP by small companies and transfer to the current business model: “fabless+foundry” – the problem is that most ReRAMs are fundamentally flawed at the scientific level to get pass the scrutiny of companies that need the product differentiation but want a long term and cost driven answer to NVM beyond FLASH. (b) Faulty science means short lifetime of products based on the “Fundamental NVM switch” for any particular technology.
    Now, what do I back this with?
    Well, the ReRAM devices shown thus far, are “chemical” reactions which involve mass transfer at the atomic level. Whether it is PCM,CBRAM,electrolytic ReRAMs, etc.It has been more than 10 years that filaments, the mainline “answer” for ReRAMs and “Vacancy-driven” mass transfer, are researched and have caused new start ups to seek funds and try to prove that the “secret sauce” works. Unfortunately, such a path is not able to convince “foundry-based” high end fabless companies to make a convincing argument to the manufacturer to pursue complicated and “limited” ReRAM “solutions”.There is only one answer, I believe: show something different and based on non-mass transfer centric devices. One way is of course, use the Mott transition, but the lack of knowledge is clear when only the Mott-transistor is talked about, and that is like fixing slide rules to solve the calculator competition. The new switch has to be electronic or it will die a slow death.Soon someone will ask me to actually show what I mean…so far, this is seen as another “secret sauce”, but it is way beyond that, it is one of those gestalt things that some see the sun going around the earth and others see the earth going around the sun. Models matter. Technology without science is like bad coffee, it is coffee but the taste is terrible.

    1. admin

      Hi Carlos, Good point, Early stage IP is still of considerable value in the ReRAM field and the IP itself could easily be of sufficient value in itself to enable the next step down the funding route. I also wholeheartedly agree that a memory where atoms do not move is almost certainly going to have a better endurance that a memory where there is no atomic displacement. DRAM is the perfect example. Please let us know when you can say more!
      Best Regards, Christie

  2. don hilliard

    Ford Maverick – that’s just great – love the juxtaposition.

    I think it’s good and much in need to point out the pure, inescapable intertial issues involved when people start getting starry-eyed about any new NVM approach.

    In response to the objection – I agree that mass transfer is an inherently high-risk approach. So, let’s looks at phase-transitions, and similar switching approaches invoking a stress-strain modulation. Like ferroelectric switching! Oh, wait.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>