Static information storage and retrieval – Read only systems – Resistive
Patent
1994-09-19
1996-07-30
Nguyen, Viet Q.
Static information storage and retrieval
Read only systems
Resistive
365 46, 365148, G06F 1580, G11C 2700
Patent
active
055418694
DESCRIPTION:
BRIEF SUMMARY
BACKGROUND
1. Field of the Invention
This invention relates to resistive memory elements suitable for use as synaptic elements in neural networks.
2. Related Art and Other Considerations
Resistive memory elements usually take the form of a resistive layer which is located between two electrodes, e.g. metallic electrodes, which give access to the element. The resistive layer is usually formed of a host material, e.g. amorphous silicon, which contains a dopant to improve its host properties. In addition, the memory element includes an active species which plays a substantial part in the conduction of an electric current and adjusts the resistance of the resistive layer. The active species is usually a metal dispersed in the host as small particles or as atoms.
(Synaptic elements are usually configured as arrays in which many elements, e.g. 10.sup.3 -10.sup.8, are supported on a single substrate. The substrate provides mechanical support for all of the elements, but the substrate has little or no effect upon the preformance of the elements.)
Each element may have its own separate electrodes, so that arrays can be connected into networks as desired. It is also possible that any one element may share one or both of its electrodes with other elements. This means that the elements are permanently connected into particular network patterns. Arrays of elements are conveniently manufactured by depositing the various layers on a suitable substrate. The layers may be patterned by using well established techniques such as masking and etching.
It has been mentioned above that the resistive layer is usually formed of a host material which includes an active species. The elements V, Co, Ni, Pd, Fe and Mn are particularly suitable as the active species for use with amorphous silicon as the host material. The active species is conveniently introduced into a memory element by making one electrode out of the active species and causing it to migrate into the host. The process of migration is "forming" and it comprises applying a high voltage, e.g. of the order 8-35 volts, for a short time, e.g. 200-1,000 ns. The forming process reduces the resistance of the element from a very high value (in practice an infinite value) to a value which is suitable for use in a neural network, e.g. to a value within the range 10.sup.3 -10.sup.6 .OMEGA..
It is emphasized that forming introduce the active species into the host material to give an initial value to the resistance. After forming, the resistance can be adjusted to a target value, and it is possible for the resistance to be adjusted and readjusted many times during the life of an element. The resistance of the element is used as the memory parameter. Resistive elements are used as the programmable memories in neural networks. The circuitry of a neural network addresses a particular element by applying an address voltage to the element, whereby the resistance of the element produces a particular current.
It is important to recognize that each element has two threshold voltages. The lower threshold voltage represents the upper limit for the address voltage. In other words the element is addressed by applying a voltage less than the lower threshold voltage to it. The upper threshold voltage constitutes the value at which the forming process may occur. Thus a voltage in excess of the upper threshold voltage is normally applied to each element in order to form it and, thereafter it is undesirable to apply voltages above the upper threshold value. Voltages which lie between the upper and lower threshold values are conveniently called "programming" voltages because voltages in this range are utilized to adjust the resistance of the element. Since the resistance of the element constitutes the memory parameter, voltages between the upper and lower threshold values are utilized for a process equivalent to the storage of data. (The data stored in its synaptic elements is an important part of the program of a neural network.)
It has been proposed to adjust the resistance of a synaptic element by
REFERENCES:
patent: 4876668 (1989-10-01), Thakoor et al.
patent: 4931763 (1990-06-01), Ramesham et al.
patent: 5335219 (1994-08-01), Ovshinsky et al.
patent: 5406509 (1995-04-01), Ovshinsky et al.
patent: 5414271 (1995-05-01), Ovshinsky et al.
Electronic Letters, vol. 24, No. 19, 15 Sep. 1988, Shimabukuro et al., "Dual-Plurality Nonvolatile MOS Analog Memory (MAM) Cell For Neural-Type Circuitry", pp. 1231-1232.
Hajto Janos
Le Comber, deceased Peter G.
Osborne Ian S.
Owen Alan E.
Rose Mervyn J.
British Telecommunications plc
Nguyen Viet Q.
LandOfFree
Resistive memory element does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Resistive memory element, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Resistive memory element will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-1665576