Feb 10

Forecasting and Filaments

YoleA couple of items have come my way recently. Many thanks to those who have alerted me to them. First, Yole have just released a new technology and market analysis dedicated to the Emerging Non Volatile Memory. The authors from Yole point out that “The technological choice between STTMRAM and RRAM will be made in the next two years: what is the best solution for data centers and embedded MCU applications?” (quotes from Yole press release) In the report “Yole presents an overview of the semiconductor memory market including NAND, DRAM, embedded MCU and mobile CPU ……..(Yole) details the emerging NVM current and future applications and emerging NVM market forecasts, provides technical roadmap of each memory, and recent developments of both dominant embedded and stand-alone memory players…” A preview is available along with details of ordering the full report. Yole point out that one of the post NAND technologies appears to have already fallen by the wayside as “In 2014, Micron, the main phase change memory (PCM) promoter for stand-alone memory, stopped actively selling PCM chips following the collapse of sales targeting the shrinking entry-level mobile phone market.” For embedded PCM will “2015 will be a key year as STMicroelectronics, the main PCM promoter in this market, will choose if PCM will remain in its roadmap.” Interesting!

The decisions behind the future choice between STTMRAM/MRAM and ReRAM for various applications is a theme of the report. For example, “In the long term, STTMRAM is sure to be the only candidate to substitute DRAM thanks to its high endurance. RRAM is sure to substitute NAND thanks to its high scalability/low cost.” But before then, “STTMRAM/MRAM and RRAM have different features and positioning. Nevertheless, they will compete in 2015 and 2016 in some standalone markets, with storage class memory for enterprise storage being the biggest one. They will also compete in embedded MCU markets in the wearable, smart card and other markets. Micron has already selected RRAM for 2015 and other key stand-alone players like Samsung and SK Hynix should react quickly. In the embedded memory space, only Panasonic has selected RRAM and many key players have not yet made their choice.” These trends are summarized in the figure below as the opportunities for ReRAM adoption develop starting with embedded memory.
Emerging NVM Technological Choices

Chinese Academy of SciencesChanging gears somewhat, a paper entitled ‘Evolution of conductive filament and its impact on reliability issues in oxide-electrolyte based resistive random access memory’ by a group from the Lab of Nano-fabrication and Novel Devices Integration, Institute of Microelectronics, Chinese Academy of Sciences in Beijing has just appeared in Nature’s online scientific reports. The authors have studied and performed failure analysis on a 1k 1T1R array of a Cu/HfO2/Pt electrochemical metallization cells (ECM) and “…. reveal the degradation of high resistance state behaves as the majority cases of the endurance failure of the….. ECM cell. High resolution transmission electron microscopy was used to characterize the change in filament nature after repetitive switching cycles. The result showed that Cu accumulation inside the filament played a dominant role in switching failure….”

It is an interesting and thorough paper although I must confess I am always a little concerned when conclusions appear to be drawn based on a couple of TEM images. I’m sure the authors must have taken many more but they have chosen not to include them even in the supplementary information. The images represent part of a circular Cu via where there is a protuberance and show Cu build up in the electrolyte above. (I did wonder how they managed to find this critical part of the contact when preparing the TEM sample) It is not clear if the contact/cell shown is typical of the entire array (in which case it looks as if the CMP process could be improved).

Many thanks Chakku and Sandrine for alerting me to these two publications!
Christie Marrian, Director ReRAM Forum

1 comment

  1. Carlos A. Paz de Araujo

    The Chinese Academy paper, and other similar papers, show that the HRS is the focus of the reliability problem. In fact, this is on of the most careful paper I have seen. It shows a lot of electrical data, and not just TEM pictures. The temperature effect in the “electromigration” of Cu from the electrodes is very clear and it starts even a room temperature. The activation energy is very low.
    This is the key weakness in RRAM/ReRAMs as is exemplified in this paper: mass transport due to filaments. Some other devices, establish the filament and then stop growth, depending only on electrochemical reactions at the anode, to close or open the direct short to the electrode. Thus, the locus of switching between two states is pretty much in a very complicated and damaged region.
    On the other hand, STTRAM is extremely difficult to also achieve reliability in a very high node, like 28 nm. The main reason is the cross product magnetization dependence in area, and the complexity of the process/design. So it is clear that we keep seeing the best data and not the large disparity in the data.
    We have demonstrated that CeRAM has perfectly flat HRS state to 400 C. And, unfortunately, in your blog, the “correlated electron” which appears in PrSrMnO, which is not useful since it is correlation based on spin-orbital coupling, is the only discussion that was presented. Strongly correlated electron, in the Mott transition, is in effect the dominant oxidation/reduction reaction in all the filament connect/disconnect ReRAM (not in the case of mass transfer CBRAM type of devices). So, qualitatively, almost all ReRAMs including the outliers in the HfO and Ta2O3 devices with TiO, CuO TMOs in a Sandwich, are in a sense using the “disproportionation reaction” one way or another in the anode locus. So, the ability to design a device, excluding filaments is key for the success of ReRAM, and the answer is to make use of the Bulk, not to create filaments, but to maintain the Strongly correlated electron phenomena in the Mott-like Metal/insulator transition throughout and not just the surface as an accident, as in most ReRAMs. This is the central point in CeRAM, which is fundamentally a very different device with distant device physics and technology control than what has been done up to now. Unfortunately, the comparison of RRAMs to STTRAM is being done in sheer ignorance of what is going on in CeRAM (mostly because very little makes the news, for proprietary reasons), but something has to be done not to cast RRAMs into a black hole and allow STTRAMs to become the next big nothing as PCM ended up being,

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>