Numonyx's PCM Truth: A Revolution That Has Not Happened

    Anecdote instead of the foreword:
    - You heard that Ivanov won the Volga lottery.
    - Actually, not Ivanov, but Rabinovich, not the Volga, but a hundred rubles, not in the lottery, but in the preference, and did not win, but lost.

    A link to the stunning news appeared recently in Habré : a revolution has occurred in the technology of memory production, and soon we will have a single memory super-duper memory based on chalcogenide materials, using phase transitions caused by heating when an electric current is passed through the cell. It works quickly like DRAM, and even non-volatile like Flash.

    Unfortunately, almost everything written in the article is wrong ...
    So that the habrasociety does not have a misconception about what is happening in the world of memory, I would like to tell the true story of PCM.

    To a perfectly reasonable question from the gallery: “And who are you so that you supposedly listen to your truth to us?” I can answer that I work in the development department of a memory manufacturing company (including PCM) and, as a result of my service, “I keep my finger on the pulse”.

    I will not comment on the points of nonsense and factual errors by reference, there are too many of them. In one sentence: PCM is a stillborn technology, there will be no revolution (and never has).

    The idea of ​​PCM is old (patent obtained in 1966), from a scientific point of view everything is clear and simple there. Some materials (chalcogenide glasses, for example) have two quasistable states: crystalline and amorphous. In crystalline, they conduct current, but in amorphous - not very. Therefore, by measuring the resistance of such a cell, information can be read.
    How to write / erase it? Everything is quite simple: if the chalcogenide is heated to a high temperature (but below the melting point), it will go into an amorphous state. If heated even more, it will melt, but upon subsequent cooling it will recrystallize into a conducting phase.

    Even about 7 years ago there was some boom in research (not in the academy, but in the industry), due to the fact that there was a real opportunity to put the technology into mass production. Several companies licensed it, many conducted research. Then it really seemed to many that PCM was the future.
    About 3 years ago, it became clear that nothing good will come of it: it is too expensive to manufacture, materials are very uncomfortable, problems with uneven recrystallization. If the cells are made small enough (and nobody needs large ones), then the resistances of the unsuccessfully recrystallized cell and its neighbor in the amorphous state will be almost the same. The main problem: thermal insulation of cells - it is difficult to hide from heat in a small volume, and when switching some cells, neighboring ones gradually lose crystallinity.

    At the same time, a breakthrough occurred in the flash memory market: multilevel cells appeared. When writing / reading from them, more than two signal levels can be distinguished, and, therefore, two, three, and in the long term, four bits can be stored in one physical cell. In addition, many miniaturization problems were resolved: now mass production includes flash memory made using a process technology of about 20 nm.

    However, a couple of companies (the first was Numonyx) released chips: just to minimize losses and not lose face. Since 2008, you can easily buy PCM chips, only almost no orders. The recording density (Numonyx chips have a capacity of 128Mbit) is too low even for mobile applications. Samsung, with some delay, released 512Mbit chips made using more advanced technology. But the lineup of buyers is also not observed.

    It’s not planned to complete the technology (miniaturization below 60nm) in the near future, in our company, for example, 5 years ago there was a whole department that dealt with PCM, today it is one person (rather stupid, you just had to stick it somewhere) . The only chance (in my opinion): “implanting” a small amount of PCM into flash drive controllers (for caching), but DRAM is usually costed there.

    Where does the legs of this homegrown sensation come from (and what does the joke from the beginning of the article have to do with it)?
    Many sites have recently written about the new Numonyx PCM chips. Basically, these are sets of general words about prospects, such as the one to which the original topic refers. They were written, for the most part, simply by journalists "on the iron", which, unfortunately, are often completely off topic.

    News onthe engineering resource sounds like this: “Numonyx announces a delay in the release of 1Gbit PCM memory .
    In fact, they cannot even repeat the result of Samsung ... But to maintain a reputation (Numonix is ​​in a difficult situation, the former founders of STM and Intel are selling the company to Mikron), the marketers came up with the idea of ​​renaming the memory two years ago with a new name and convening a press conference from journalists “not in topic ”to create an information buzz about their success.

    I am sure that in a few years everyone will calm down and stop releasing PCM, which, unfortunately, turned out to be a stillborn product.

    Also popular now: