Memresistor (i.e. memory resistor) electronics is defined by materials which have a memory of previously applied voltages or currents. Since the late 1990’s many different companies including Samsung, Micron Technology, Sharp, Hewlett- Packard, and Unity Semiconductor have been working with phase change, solid electrolyte, and metal oxide materials having memresistor characteristics to produce “transistor-less” memory cells or to introduce new concepts in computing and neuromorphics. This article reviews the patents and companies related to the memresistor concept.
A spreadsheet containing the patent data used for this article is available by sending an E-mail to firstname.lastname@example.org including the subject “memresistor patents.” For information on why HP’s memristor might be considered scientific fraud the reader is referred to this link. Some of the memresistor patents issued by the USPTO appear invalid and I created an online database of prior art citations filed under 37 CFR 1.501 available at this link.
Over the past 40 years semiconductor electronics has been driven by Moore’s Law which predicts an exponential reduction in the size of transistors over time providing more functionality for electronic devices at less cost. However, due to physical limitations of scalability and manufacturing cost, Moore’s Law cannot continue forever and the optimization of conventional silicon transistors will inevitably come to an end. The Semiconductor Industry Roadmap forecasts further improvements in the next few years but past 2012 the future of electronics is beginning to appear uncertain. Furthermore, device electronics are evolving to require more adaptability in function such as cell phones incorporating the functions of internet, music, and video provider and video game interfaces including voice and image recognition capabilities. The degree of adaptability required to achieve these functions is gradually moving beyond what conventional computer architectures can provide. It is becoming clear that new materials other than silicon and new electronics architectures based on such new materials will be necessary to provide continued advancement in electronic functionality over the next few decades. Memresistors (memory resistors) may be the key to such new materials and architectures. Like the transistor, memresistors behave as a switch but, unlike the transistor, the mem-resistor is a 2-terminal rather than a 3-terminal device. A comparison of the attributes of the transistor vs. the memresistor is given by TABLE 1 below. A listing of the top U.S. patent holders for memresistors is given in TABLE 2 (based on first listed assignee).
|3-terminal switching device with an input electrode (e.g. source), an output electrode (e.g. drain), and a control electrode (e.g. gate)||2-terminal device with one of the electrodes acting either as a control electrode or a source electrode depending on the voltage magnitude.|
|Requires a power source to retain a data state||Does not require a power source to retain a data state|
|Stores data by electron charge||Stores data by resistance state|
|Scalable by reducing the lateral length and width dimensions of channel||Scalable by 3D stacking|
|Capable of performing analog or digital electronic functions depending on applied bias voltages||Capable of performing analog or digital electronic functions depending on previous voltage history|
|Fabrication requires optical lithography||Fabrication by optical lithography but alternative (potentially cheaper) production techniques such as nanoimprint lithography and self assembly are possible|
|Patent Holder(Total Patents)||’98||’99||’00||’01||’02||’03||’04||’05||’06||’07||’08||’09||’10||’11|
|Unity Semi (86)||5||12||21||5||13||5||8||17|
|Hynix Semi (71)||1||4||3||3||4||22||34|
|Sandisk 3D (66)||3||8||6||27||22|
|Seagate Tech (57)||15||42|
|Industrial Technology ResearchInstitute (43)||2||2||7||19||13|
The three main areas of application covered by the patents for memory resistance electronics are non-volatile memory, logic/computation, and neuromorphics. Table 2 provides data on the issued U.S. patent count among the leading patent holders in each area. The category “other” designates patents covering memresistor manufacture or applications in other areas (e.g. analog electronics).
|Patent Holder (Total Patents)||Memory||Logic/Computation||Neuromorphics||Other|
|Unity Semi (86)||78||1||7|
|SanDisk 3D (66)||58||8|
|Industrial Technology Research (43)||38||5|
|Energy Conversion Devices (43)||25||6||2||10|
As evident from the patent data, non-volatile memory is the dominant area being pursued for memresistor technology. Of course most of the companies listed do not refer to their memory in terms of a memresistor and rather use a variety of acronyms (i.e. RRAM or ReRAM, CBRAM, PRAM, etc.) to distinguish their particular memory design. HP has attracted much attention in the popular science press since 2008 by associating their version of a memresistor with a theoretical circuit element called a “memristor”originally defined by Leon Chua in 1971. However, Chua’s memristor is not a scientifically accurate concept and lacks generality (which is why I instead use the term memresistor).
Flash memory currently dominates the semiconductor memory market. However, each memory cell of flash requires at least one transistor meaning that flash design is highly susceptible to an end to Moore’s law. On the other hand, memresistor memory design is often based on a crossbar architecture which does not require transistors in the memory cells. Although transistors are still necessary for the read/write circuitry, the total number of transistors for a million memory cells can be on the order of thousands instead of millions and the potential for addressing trillions of memory cells exists using only millions (instead of trillions) of transistors. Another fundamental limitation to conventional memory architectures is Von Neumann’s bottleneck which makes it more difficult to locate information as memory density increases. Memresistors offer a way to overcome this hurdle since they can integrate memory and processing functions in a common circuit architecture providing a de-segregation between processing circuitry and data storage circuitry.
HP is collaborating with Hynix Semiconductor and has projected that they would be producing a metal oxide memresistor memory in 2014 or 2015. Unity Semiconductor was bought by Rambus in February 2012 for $35 million. Unity was previously reported to have a 2-year joint development program with Micron Technology. Sharp and Elpida have projected their version of ReRAM to be ready by 2013. Panasonicreleased the first ReRAM prototypes in 2012.
The uses of memresistor technology for logic and computational electronics is less well developed than for memory architectures but the seeds of innovation in this area are currently being sown. Memresistors appear particularly important to the areas of reconfigurable computing architectures such as FPGAs in which the arrangement between arrays of basic logic gates can be altered by reprogramming the wiring interconnections. Memresistors may be ideal to improve the integration density and reconfigurability of such systems. In addition, since some memresistor materials are capable of tunablity in their resistance state they can provide new types of analog computational systems which may find uses in modeling probabilistic systems (e.g. weather, stock market, biosystems) more efficiently than purely binary logic-based processors. Hewlett-Packard researchers have developed some ideas in using memresistors for reconfigurable computing although Stanford Ovshinsky (formerly of Energy Conversion Devices) earlier developed the key ideas on using the analog properties of phase change materials for computation.
Neuromorphics has been defined in terms of electronic analog circuits that mimic neuro-biological architectures. The analog properties of memresistors may provide the ideal device characteristics for neuromorphics. In the 1960’s Bernard Widrow used a 3-terminal variation of a memresistor called a memistor to implement some of the earliest versions of single-layer and multi-layer artificial neural networks. While so far few patents have been issued involving memresistors applied to neuromorphics, DARPA initiated the SyNAPSE project in 2009 funding IBM, HRL, and HP in the development of neuromorphic architectures.
Since the 1960’s a wide variety of materials have been found to possess memory resistance properties. The four main categories which have been the areas of focus over the past decade are phase change, solid electrolyte/ionic, metal oxide, and molecular/polymer. TABLE 4 provides data on the distribution of the patents among the top assignees. It is notable that there is some overlap between the four main categories. For example, metal oxides might be considered ionic based on charge carriers such as oxygen vacancies. Also chalcogenide material is used to form both phase change and solid electrolyte memresistors. The categorization used in determining the type is based on how the claims of each patent described the material. It is also noted that the types of materials listed are not the only materials exhibiting memory resistive effects and some newly discovered memory resistive materials are based on amorphous silicon, nanoparticle assemblies, and hydrogel materials.
|Solid Electrolyte||Metal Oxide||Molecular/ Polymer||Other/ Unspecified|
|Unity Semi (86)||2||18||66|
|Hynix Semi (71)||62||1||2||6|
|SanDisk 3D (66)||4||9||53|
Research Inst (43)
|Energy Conversion Devices (43)||26||17|