Patent Grant
8363446
This invention relates to the operation of variable resistance memory devices. More particularly, this invention relates to a method of programming that reduces the deleterious effect of drift on the stability of the resistance states of phase-change materials in multilevel memory devices. Most particularly, this invention relates to a method of programming a phase-change material in a multilevel device that utilizes crystalline programming states having low drift.
Variable resistance materials are promising active materials for next-generation electronic storage and computing devices. A variable resistance material is a material that possesses two or more states that differ in electrical resistance. The material can be programmed back and forth between the states by providing energy to induce an internal chemical, electronic, or physical transformation of the material that manifests itself as a change in resistance of the material. The different resistance states are distinguishable and can be used as memory states to store or process data.
Phase change materials are a promising class of variable resistance materials. A phase change material is a material that is capable of undergoing a transformation, preferably reversible, between two or more distinct structural states. The distinct structural states may be distinguished on the basis of, for example, crystal structure, atomic arrangement, order or disorder, fractional crystallinity, relative proportions of two or more different structural states, or a physical (e.g. electrical, optical, magnetic, mechanical) or chemical property. In a common embodiment, the two or more distinct structural states include differing proportions of crystalline phase regions and amorphous phase regions of the phase change material, where the phase-change material is reversibly transformable between the different states. In the crystalline state, the phase change material has lower resistivity; while in the amorphous state, it has higher resistivity. Continuous variations in resistivity over a wide range can be achieved through control of the relative proportions of crystalline phase regions and amorphous phase regions in a volume of phase-change material. Reversibility of the transformations between structural states permits reuse of the material over multiple cycles of operation.
Typically, a programmable resistance device is fabricated by placing the active variable resistance material, such as a phase change material, between two electrodes. Operation of the device is effected by providing an electrical signal between the two electrodes and across the active material. In a common application, phase-change materials may be used as the active material of a memory device, where distinct data values are associated with the different structural states and each data value corresponds to a distinct resistance range of the phase-change material. The different structural states employed in memory operation may also be referred to herein as memory states or resistance states of the phase-change material. Write operations in a phase-change memory device, which may also be referred to herein as programming operations, apply electric pulses to the phase-change material to alter its structural state to a state having the resistance associated with the intended data value. Read operations are performed by providing current or voltage signals across the two electrodes to measure the resistance. The energy of the read signal is sufficiently low to prevent disturbance of the structural state of the phase-change material.
Phase-change memory devices are today typically operated in binary mode. In binary mode, the memory is operated between two structural states. To improve read margin and minimize read error, the two structural states for binary operation are selected to provide a large resistance contrast. The range of resistance values of a phase-change material is bounded by a set state having a set resistance and a reset state having a reset resistance. The set state is a low resistance structural state whose electrical properties are primarily controlled by the crystalline portion of the phase-change material and the reset state is a high resistance structural state whose electrical properties are primarily controlled by the amorphous portion of the phase-change material. The set state and reset state are commonly employed in binary operation and may be associated with the conventional binary “0” and “1” states.
In order to expand the commercial opportunities for phase-change memory, it is desirable to identify new phase-change compositions, device structures, and methods of programming that lead to improved performance. A key performance metric for memory devices is storage density, which is a measure of the amount of information that can be stored per unit area of memory material. Miniaturization is the most common strategy for increasing storage density. By shrinking the area required to store a bit of information, more bits can be stored in a memory chip of a given size. Miniaturization has been a successful strategy for increasing storage density over the past four decades, but is becoming increasingly more difficult to employ as fundamental size limits of manufacturability are reached.
An alternative approach for increasing storage density is to increase the number of bits stored in a given area of memory. Instead of reducing the area in which information is stored, the objective is to store more bits of information in a particular area of memory. In conventional binary operation, only a single bit of information is stored in each memory location. Higher storage density can be achieved by increasing the storage capacity of each memory location. If two bits, for example, can be stored at each memory location, the storage capacity doubles without miniaturizing the memory location. In order to increase the storage capacity of each memory location, it is necessary for the memory material to be operable over more than the two states used in binary (single bit) operation. Two-bit operation, for example, requires a material that is operable over four distinguishable memory states.
Phase-change memory materials have the potential to provide multiple bit operation because of the wide resistance range that separates the set and reset states. In a typical phase-change memory device, the resistance of the set state is on the order of ˜1-10 kΩ, while the resistance of the reset state is on the order of ˜100-1000 kΩ. Since the structural states of a phase-change material are essentially continuously variable over the range of proportions of crystalline and amorphous phase volume fractions extending from the set state to the reset state, memory operation of a phase-change material at memory states having resistances intermediate between the set resistance and reset resistance is possible. As a result, multiple bit memory operation over multiple memory states is achievable with phase-change memory materials.
Although phase-change memory offers the potential for multiple bit operation, progress toward achieving a practical multilevel phase-change memory has been limited. One of the practical complications associated with multilevel phase-change operation is resistance drift over time. It is common in phase-change memory devices to observe a variation in the resistance of a memory state over time. If a phase-change memory device is programmed into a particular state having a particular resistance at one time, the resistance of the device at a later time may be different. As a general rule, resistance increases with time and becomes more pronounced as the amorphous phase volume fraction of a structural state increases. Resistance drift is not problematic for binary operation of phase-change memory because the set state shows little or no drift in resistance over time, while the reset state shows an increase in resistance over time. As a result, the resistance contrast between the set and reset states increases over time and no impairment of performance occurs.
Resistance drift, however, becomes problematic in multilevel applications of phase-change memory devices because time variations in resistance may lead to overlap in the identification of memory states. In order to advance the performance capabilities and commercial potential of phase-change memory, it is desirable to develop phase-change materials, device structures or methods of operating phase-change memory devices that eliminate, counteract, or circumvent resistance drift.
This invention provides a method for programming phase-change memory in a multilevel mode of operation that diminishes the practical effect of resistance drift on performance. The programming method recognizes that the intrinsic drift of crystalline phase states of a phase-change material is much less pronounced than the intrinsic drift of amorphous phase states. The method accordingly provides a programming strategy that utilizes two or more crystalline programming states based on different crystalline phases.
The method generally permits programming between three or more states that are resolvable on the basis of resistance where at least two of the programming states are crystalline states. In one embodiment, the method utilizes two or more crystalline states and an amorphous or partially amorphous state as programming states. The two or more crystalline programming states are distinguishable on the basis of resistance and exhibit little or no drift in resistance over time. The amorphous or partially amorphous programming state has a much higher resistance than any of the crystalline programming states and may exhibit a drift in resistance over time. The resistance drift of the amorphous or partially amorphous state does not, however, lead to ambiguity in identifying programming states. In one embodiment, drift causes the resistance of the amorphous or partially amorphous state to increase in time at a rate that increases the resistance contrast between the amorphous or partially amorphous state and any of the crystalline programming states. As a result, overlapping of states is avoided.
In one embodiment, the programming method utilizes at least two crystalline programming states and one amorphous programming state. In another embodiment, the programming method utilizes at least two crystalline programming states and a mixed crystalline-amorphous programming state. In still another embodiment, the programming method utilizes at least three crystalline programming states. Additional crystalline, mixed crystalline-amorphous, and amorphous states may be included in the method along with the at least two crystalline states to obtain additional programming states. In one embodiment, the method includes four or more programming states. In another embodiment, the method includes six or more programming states. In a further embodiment, the method includes eight or more programming states.
The crystalline programming states may be different crystalline phases of a phase-change material. In one embodiment, the at least two crystalline programming states include one state that is predominantly face-centered cubic and a second state that is predominantly hexagonal, where the resistances of the face-centered cubic and hexagonal states are distinguishable. In another embodiment, the at least two crystalline programming states include one state that is predominantly face-centered cubic and one state that is a mixture of face-centered cubic and hexagonal phases. In still another embodiment, the at least two crystalline programming states include one state that is predominantly hexagonal and one state that is a mixture of face-centered cubic and hexagonal phases. In a further embodiment, the at least two crystalline programming states include one state that is predominantly face-centered cubic, one state that is a mixture of face-centered cubic and hexagonal phases, and one state that is predominantly hexagonal. Each of the foregoing may further include one or more amorphous or partially amorphous programming states.
Although this invention will be described in terms of certain preferred embodiments, other embodiments that are apparent to those of ordinary skill in the art, including embodiments that do not provide all of the benefits and features set forth herein, are also within the scope of this invention. Accordingly, the scope of the invention is defined only by reference to the appended claims.
The instant invention is directed at enabling multilevel operation of phase-change memory devices by selecting programming states that diminish the negative consequences associated with resistance drift. In order to appreciate the benefits of the instant invention, it is helpful to review the basic operational characteristics of phase-change memory materials and devices and to discuss the problem of resistance drift. The following discussion focuses on chalcogenide materials as illustrative phase-change materials. The basic principles apply equally to other forms of phase-change, state-change or variable resistance materials, such as pnictides or other classes of materials transformable between two or more states distinguishable on the basis of structure, physical property or chemical property.
An important feature of the chalcogenide materials in the operation of chalcogenide-based phase-change memory devices and arrays is their ability to undergo a phase transformation between or among two or more structural states. The chalcogenide materials have structural states that may include a crystalline state, one or more partially-crystalline states and an amorphous state. The crystalline state may be a single crystalline state or a polycrystalline state. A partially-crystalline state refers to a structural state in which a volume of chalcogenide or phase-change material includes an amorphous portion and a crystalline portion. A partially-crystalline state may also be referred to herein as a mixed crystalline-amorphous state. Generally, a plurality of partially-crystalline states exists for the chalcogenide or phase-change material that may be distinguished on the basis of the relative proportion of the amorphous and crystalline portions. Fractional crystallinity is one way to characterize the structural states of a chalcogenide phase-change material. The fractional crystallinity of the crystalline state is 100%, the fractional crystallinity of the amorphous state is 0%, and the fractional crystallinities of the partially-crystalline states may vary continuously between 0% (the amorphous limit) and 100% (the crystalline limit). Phase-change chalcogenide materials are thus able to transform among a plurality of structural states that vary inclusively between fractional crystallinities of 0% and 100%.
Transformations among the structural states are induced by providing energy to the chalcogenide material. Energy in various forms can induce structural transformations of the crystalline and amorphous portions and thus can influence the fractional crystallinity of a chalcogenide material. Suitable forms of energy include one or more of electrical energy, thermal energy, optical energy or other forms of energy (e.g. particle-beam energy) that induce electrical, thermal or optical effects in a chalcogenide material. Continuous and reversible variability of the fractional crystallinity is achievable by controlling the energy environment of a chalcogenide material. A crystalline state can be transformed to a partially-crystalline or an amorphous state, a partially-crystalline state can be transformed to a crystalline, amorphous or different partially-crystalline state, and an amorphous state can be transformed to a partially-crystalline or crystalline state through proper control of the energy environment of a chalcogenide material. Some considerations associated with the use of thermal, electrical and optical energy to induce structural transformations are presented in the following discussion.
The use of thermal energy to induce structural transformations exploits the thermodynamics and kinetics associated with the crystalline to amorphous or amorphous to crystalline phase transitions. An amorphous phase may be formed, for example, from a partially-crystalline or crystalline state by heating a chalcogenide material above its melting temperature and cooling at a rate sufficient to inhibit the formation of crystalline phases. A crystalline or higher fractional crystallinity phase may be formed from an amorphous or partially-crystalline state, for example, by heating a chalcogenide material above the crystallization temperature for a sufficient period of time to effect nucleation and/or growth of crystalline domains. The crystallization temperature is below the melting temperature and corresponds to the minimum temperature at which crystallization may occur in a practical amount of time. The driving force for crystallization is typically thermodynamic in that the free energy of a crystalline or partially-crystalline state in many chalcogenide materials is lower than the free energy of an amorphous state so that the overall energy of a chalcogenide material decreases as the fractional crystallinity increases. Formation (nucleation and growth) of a crystalline state or crystalline domains within a partially-crystalline or amorphous state is kinetically enabled up to the melting temperature, so that heating promotes crystallization by providing energy that facilitates the rearrangements of atoms needed to form crystalline phases or domains.
The fractional crystallinity of a partially-crystalline state can be controlled by controlling the temperature or time of heating of the partially-crystalline state or by controlling the temperature or rate of cooling of an amorphous or partially-crystalline state. Through proper control of the peak temperature, time of heating and rate of cooling, structural states over the full range of fractional crystallinity can be achieved for the chalcogenide phase-change materials.
The use of electrical energy to induce structural transformations relies on the application of electrical (current or voltage) pulses to a chalcogenide material. The mechanism of electrically-induced structural transformations includes Joule heating created by resistance to current flow. Joule heating corresponds to a conversion of electrical energy to thermal energy and leads to an increase in the temperature of the chalcogenide material. By controlling the current density, the temperature can be controlled.
The crystalline phase portions of a chalcogenide material are sufficiently conductive to permit current densities that provide efficient Joule heating. The amorphous phase portions, however, are much less conductive and ordinarily would not support current densities sufficient to heat the material to the crystallization temperature. As described more fully hereinbelow, however, the amorphous phase of many chalcogenide materials can be electrically switched to a highly conductive intermediate (dynamic) state upon application of a voltage greater than the threshold voltage. In the intermediate state, the material can support a current density that is high enough to heat the material to the crystallization temperature through Joule heating. By controlling the magnitude and/or duration of electrical pulses applied to a chalcogenide phase-change material, it is possible to continuously vary the fractional crystallinity through controlled interconversion of the crystalline and amorphous phases.
The influence of electrical energy on a chalcogenide material is generally depicted in terms of the I-V (current-voltage) and R-I (resistance-current) relationships of the material. The I-V relationship shows the current response of a chalcogenide material as a function of applied voltage and the R-I relationship shows the variation of the electrical resistance of a chalcogenide material as a result of applying electrical energy or as a function of the magnitude of the current or voltage pulse applied to a chalcogenide material. The resistance plotted on the R-I plot is the resistance measured at a fixed time (t0) after the application of the current pulse (I) and is measured at a sufficiently low voltage as to not disturb the memory state. A brief discussion of the I-V and R-I characteristics of chalcogenide materials follows.
The I-V response of many chalcogenide materials exhibits an electrical switching event in which the chalcogenide material undergoes a transformation from a more resistive state to a more conductive state. When present, an electrical switching event is generally observed only in amorphous phase chalcogenide materials or mixed crystalline-amorphous phase chalcogenide materials and is ordinarily not observed in crystalline phase chalcogenide materials. As discussed more fully hereinbelow, the limiting crystalline and amorphous phase structural states of a chalcogenide phase-change material are normally referred to as the “set” and “reset” states, respectively.
A schematic depiction of the electrical switching event is presented in
The I-V curve includes a resistive branch and a conductive branch. The branches are labeled in
The slopes of the resistive and conductive branches shown in
The switching properties of the chalcogenide material are shown in
When the applied voltage equals or exceeds the threshold voltage, the chalcogenide material transforms (switches) from the resistive branch to the conductive branch of the I-V curve. The switching event occurs rapidly and is depicted by the dashed line in
The switching effect of the instant devices originates from a transformation of the chalcogenide material from a resistive state to a conductive state upon application of a threshold voltage, Vth. According to one model of the switching transformation, application of the threshold voltage causes the formation of a conductive channel or filament within the chalcogenide material. At the threshold voltage, the electric field experienced by the material is sufficiently high to induce a breakdown or avalanche effect whereby electrons are removed from atoms to form a highly conductive, plasma-like filament of charge carriers. Rather than being bound to atoms, some electrons become unbound and highly mobile. As a result, a conductive channel or filament forms. The conductive filament constitutes a conductive volume within the otherwise resistive chalcogenide material. The conductive filament extends through the chalcogenide material between the device terminals and provides a low resistance pathway for electrical current. Portions of the chalcogenide material outside of the filament remain resistive. Since electric current traverses the path of least resistance, the presence of a conductive filament renders the chalcogenide material conductive. The creation of a conductive filament is the event that underlies the switching of the chalcogenide material from a resistive state to a conductive state.
The conductive filament is maintained between the device terminals as long as the device current remains at or above the holding current. A conductive filament is present for all points along the conductive branch, but the cross sectional area of the filament differs for different points along the conductive branch. Depending on operating conditions within the conductive branch, the filament can be narrow or wide. As the applied voltage is increased along the conductive branch, the cross section of the filament is enlarged as the applied voltage is increased. The enlarged filament indicates that a greater volume of the chalcogenide material exhibits high conductivity. When the filament reaches the dimensions of the pore, it can no longer expand and the device resistance increases (beginning at points A in
When the chalcogenide material is initially in the set state, the switching event shown in
Chalcogenide materials of many chemical compositions undergo the foregoing switching effect. Representative chalcogenide materials are those that include one or more elements from column VI of the periodic table (the chalcogen elements) and optionally one or more chemical modifiers from columns III, IV or V. One or more of S, Se, and Te are the most common chalcogen elements included in the active material of the instant devices. The chalcogen elements are characterized by divalent bonding and the presence of lone pair electrons. The divalent bonding leads to the formation of chain and ring structures upon combining chalcogen elements to form chalcogenide materials and the lone pair electrons provide a source of electrons for forming a conducting filament. Trivalent and tetravalent modifiers such as Al, Ga, In, Ge, Sn, Si, P, As and Sb enter the chain and ring structures of chalcogen elements and provide points for branching and crosslinking. The structural rigidity of chalcogenide materials depends on the extent of crosslinking and leads to a broad classification of chalcogenide materials, according to their ability to undergo crystallization or other structural rearrangements, into one of two types: threshold switching materials and phase-change (or memory) materials.
Both types of chalcogenide materials display the switching behavior shown in
Phase-change chalcogenide materials, on the contrary, are lightly crosslinked and more easily undergo full or partial crystallization. An amorphous phase-change material undergoes filament formation in the presence of a threshold voltage as described in
The R-I response is a meaningful depiction of the characteristics of the chalcogenide phase-change memory devices and provides a representation of the effect of structural transformations associated with the crystalline-amorphous phase-change process on electrical properties. A representative depiction of the electrical resistance (R) of a chalcogenide memory device as a function of electrical energy or current pulse magnitude (Energy/Current) is presented in
The resistance plot includes two characteristic response regimes of the device to electrical energy. The regimes are approximately demarcated with the vertical dashed line 10 shown in
The left side of the resistance plot is referred to as the accumulating regime because the structural state of the chalcogenide material cumulatively evolves as energy is applied. The fractional crystallinity of the structural state correlates with the total accumulation of applied energy. The leftmost point 20 corresponds to the structural state in the accumulating regime having the lowest fractional crystallinity and may be referred to as the reset state. This state may be fully amorphous or may be primarily amorphous with some degree of crystalline content. As energy is added, the chalcogenide material progresses among a plurality of partially-crystalline states with increasing fractional crystallinity along the plateau 30. Selected accumulation states (structural states in the accumulation region) are marked with squares in
Upon accumulation of a sufficient amount of applied energy, the fractional crystallinity of the chalcogenide material increases sufficiently to effect a setting transformation characterized by a dramatic decrease in electrical resistance and stabilization of the set state 40. The structural states in the accumulation regime may be referred to as accumulation states of the chalcogenide material. Structural transformations in the accumulating regime are unidirectional in the sense that they progress in the direction of increasing applied energy within the plateau region 30 and are reversible only by first driving the chalcogenide material through the set point 40 and resetting (melting and quenching the device). Once the reset state is obtained, lower amplitude current pulses can be applied and the accumulation response of the chalcogenide material can be retraced. It is possible to reversibly transform between the set and reset states over multiple cycles of operation.
While not wishing to be bound by theory, it is the believed that the addition of energy to a chalcogenide material in the accumulating regime leads to an increase in fractional crystallinity through the nucleation of new crystalline domains or growth of existing crystalline domains or a combination thereof. It is further believed that the electrical resistance varies only gradually along the plateau 30, despite the increase in fractional crystallinity, because the crystalline domains form or grow in relative isolation of each other so as to prevent the formation of a contiguous crystalline network that spans the chalcogenide material between the two device electrodes. This type of crystallization may be referred to as sub-percolation crystallization.
The setting transformation coincides with a percolation threshold in which a contiguous, interconnected crystalline network forms within the chalcogenide material and bridges the space between the two electrodes of the device. Such a network may form, for example, when crystalline domains increase sufficiently in size to impinge upon neighboring domains. Since the crystalline phase of chalcogenide materials is more conductive and less resistive than the amorphous phase, the percolation threshold corresponds to the formation of a contiguous low resistance conductive pathway through the chalcogenide material. As a result, the percolation threshold is marked by a dramatic decrease in the resistance of the chalcogenide material. The leftmost point 20 of the accumulation regime may be an amorphous state or a partially-crystalline state lacking a contiguous crystalline network. Sub-percolation crystallization commences with an initial amorphous or partially-crystalline state and progresses through a plurality of partially-crystalline states having increasingly higher fractional crystallinities until the percolation threshold is reached and the setting transformation occurs.
The regime to the right of the line 10 of
In contrast to the accumulating region, structural transformations that occur in the direct overwrite region are reversible and bi-directional. As indicated hereinabove, each state in the direct overwrite region may be identified by its resistance and an associated current pulse magnitude, where application of the associated current pulse magnitude induces changes in fractional crystallinity that produce the particular resistance state. Application of a subsequent current pulse may increase or decrease the fractional crystallinity of an existing resistance state of the chalcogenide material. If the subsequent current pulse has a higher magnitude than the pulse used to establish the existing state, the fractional crystallinity of the chalcogenide material decreases and the structural state is transformed from the existing state in the direction of the reset state along the direct overwrite resistance curve. Similarly, if the subsequent current pulse has a lower magnitude than the pulse used to establish the existing state, the fractional crystallinity of the chalcogenide material increases and the structural state is transformed from the existing state in the direction of the set state along the direct overwrite resistance curve.
The direct overwrite states of the chalcogenide material may be used to define memory states of a memory device. Most commonly, the memory devices are binary memory devices that utilize two of the direct overwrite states as memory states, where a distinct data value (e.g. “0” or “1”) is associated with each state. Each memory state thus corresponds to a distinct structural state of the chalcogenide material and readout or identification of the state can be accomplished by measuring the resistance of the material (or device) since each structural state is characterized by a distinct resistance value. The operation of transforming a chalcogenide material to the structural state associated with a particular memory state may be referred to herein as programming the chalcogenide material, writing to the chalcogenide material or storing information in the chalcogenide material.
To facilitate readout and minimize readout error, it is desirable to select the memory states of a binary memory device so that the contrast in resistance of the two states is large. Typically the set state (or a state near the set state) and the reset state (or a state near the reset state) are selected as memory states in a binary memory application. The resistance contrast depends on details such as the chemical composition of the chalcogenide, the thickness of the chalcogenide material in the device and the geometry of the device. For a layer of phase-change material having the composition Ge22Sb22Te56, a thickness of ˜600 Å, and pore diameter of below ˜0.1 μm in a typical two-terminal device structure, for example, the resistance of the reset state is ˜100-1000 kΩ and the resistance of the set state is under ˜10 kΩ. Phase-change materials in general show resistances in the range of ˜100 kΩ to ˜10,000 kΩ in the reset state and resistance of ˜0.5 kΩ to ˜50 kΩ in the set state. In the preferred phase-change materials, the resistance of the reset state is at least a factor of two, and more typically an order of magnitude or more, greater than the resistance of the set state.
This invention seeks to extend the applicability of chalcogenide or phase-change materials beyond binary (single bit) memory applications to multilevel (non-binary or multiple bit) memory applications. The storage density of a multilevel chalcogenide or phase-change memory device improves as the number of memory states increases. Multilevel memory operation can be achieved by selecting three or more states of a phase-change memory device and associating a unique data value with each. Each of the three or more states corresponds to a distinct structural state of the chalcogenide, phase-change, or variable resistance material, and may be characterized by a distinct resistance value. Two bit operation can be achieved by selecting four states to serve as memory states, three bit operation can be achieved by selecting eight states to serve as memory states, etc.
To improve the storage density in a multilevel memory device, it is desirable to operate the memory material over as many states as possible. The number of memory states is controlled by the resistance interval between the set state and reset state, the resolution limit of the resistance measurement performed during the read operation, and the stability of the resistance values. A large resistance difference between the set and reset states provides a wide dynamic range of resistance over which operation of the memory device can occur. The resolution limit of the read resistance measurement imposes a practical limit on the spacing of resistance values associated with the different memory states. The resolution limit depends on read noise and circuit sense-amp resolution and imbalance. The resistance differential between adjacent memory states must be greater than the resolution of the read resistance measurement. Stable resistance values are needed to insure that programmed resistance values do not vary in time.
In current chalcogenide phase-change memory devices, the resistance contrast between the set and reset states is large (at least an order of magnitude) and the resolution of a resistance measurement is on the order of a few or so percent of the measured resistance value. The primary practical limitation in achieving reliable multilevel operation today is variability in the resistance of the memory states over time. Time variability of resistance may be referred to herein as “resistance drift” or “drift”. Drift is a common occurrence in chalcogenide phase-change materials and causes the resistance of the material to deviate (usually increase) from its programmed value. As indicated hereinabove, storing data in a chalcogenide memory device occurs by using energy to program the chalcogenide memory material to the structural state assigned to the data value. The resistance of the structural state is a representation of the data value and different data values are distinguished on the basis of resistance. When drift occurs, the programmed resistance deviates over time from its initial value. As a result, when the memory device is read at a later time, the resistance differs from the intended value and a misread of information may occur. The variation in resistance over time is spontaneous and occurs in the absence of any post-programming electrical stimulation.
Experimentally, it has been observed that drift normally leads to an increase in resistance over time. The variation of resistance with time is commonly modeled empirically with the following power law relation:
R(t)=R0td
where R(t) is the resistance at time t, R0 is the resistance measured 1 second (or other fixed short time) following the programming pulse, t is time, and d is the drift exponent (which may also be referred to herein as a drift coefficient).
The negative ramifications of drift for multilevel memory devices can be understood from
In the case of a binary memory, where the set state (e.g. (000)) and reset state (e.g. (111)) are selected as the memory state, drift presents no practical difficulty because the drift exponent of the set state is either non-existent or so small as to preclude the possibility of confusing the set state with the reset state. (Drift of the reset state simply results in a desirable increase in read margin). As the number of memory states increases and the resistance interval between states decreases, however, drift becomes of great practical concern.
The instant invention seeks to exploit the low drift characteristics of crystalline states for multilevel memory devices. As indicated in
The programming schemes of the instant invention utilize two or more crystalline programming states. As used herein, “crystalline programming state” refers to a programming state whose electrical resistance is controlled primarily by the crystalline phase content of the phase-change material. When a device is in a crystalline programming state, the phase-change material includes a contiguous crystalline network that spans the distance between device electrodes. The prevalence of the contiguous crystalline network is sufficient enough to preclude any meaningful contribution of an amorphous phase to the measured electrical resistance of the device. Although the physical state of a phase-change material in a crystalline programming state is expected to include primarily crystalline phase content, it may also include ancillary amorphous phase content. The set state described hereinabove is an illustrative crystalline programming state.
Similarly, “amorphous programming state” refers to a programming state whose electrical resistance is controlled primarily by the amorphous phase content of the phase-change material. When a device is in an amorphous programming state, the phase-change material includes a contiguous amorphous network that spans the distance between device electrodes and precludes the existence of a contiguous crystalline network that spans the distance between device electrodes. The absence of the contiguous crystalline network is sufficient enough to preclude any substantial contribution of a crystalline phase to the measured electrical resistance of the device. Although the physical state of a phase-change material in an amorphous programming state is expected to include primarily amorphous phase content, it may also include ancillary crystalline phase content. The ancillary crystalline phase may consist of dispersed, isolated regions of crystalline material (e.g. microcrystalline regions). The reset state described hereinabove is an illustrative amorphous programming state.
A “mixed crystalline-amorphous programming state” refers to a programming state whose electrical resistance reflects appreciable contributions from both crystalline phase content and amorphous phase content. The prevalence of crystalline phase content and amorphous phase content is sufficient so that each has an appreciable influence on the measured electrical resistance of the device. The phase-change material may, but need not, include a contiguous crystalline network. If present, however, the contiguous crystalline network is insufficient in cross-section or volume fraction to support the full current load passing between electrodes. The relative proportions and/or spatial configuration of crystalline phase content and amorphous phase content in a mixed crystalline-amorphous programming state is such that the current flowing between electrodes must pass through appreciable amounts of crystalline phase content and amorphous phase content. The measured resistance of a mixed crystalline-amorphous programming state is accordingly intermediate between the resistance of a crystalline programming state and an amorphous programming state.
The instant invention provides programming schemes that utilize two or more crystalline programming states, each of which consists essentially of or includes one or more distinct crystalline phases, where each crystalline phase is characterized by a distinct crystallographic unit cell structure. As is known in the art of crystallography, the structures of crystalline solids can be classified on the basis of space groups and unit cells (e.g. cubic, trigonal, hexagonal, tetragonal, orthorhombic, monoclinic, and triclinic). The instant inventors have shown that it is possible to stabilize two or more crystalline phases of phase-change compositions through control over the programming conditions and have further shown that the different crystalline phases are resolved in resistance and exhibit little or no resistance drift. The instant invention accordingly provides programming schemes for multilevel operation that incorporate a plurality of crystalline programming states.
In co-pending U.S. patent application Ser. No. 11/032,345 (“'345 application”), one of the instant inventors showed that a phase-change material having the composition Ge2Sb—2Te5 can transition from a more resistive face-centered cubic (fcc) crystalline phase to a less resistive hexagonal crystalline phase under the influence of Joule heating. FIG. 3 from the '345 application illustrates the effect of annealing temperature on the conductivity and physical state of Ge2Sb2Te5 and is reproduced in
The method of the instant invention generally permits programming between three or more states that are resolvable on the basis of resistance where at least two of the programming states are crystalline states. In one embodiment, the method utilizes two or more crystalline states and an amorphous or partially amorphous state as programming states. The two or more crystalline programming states are distinguishable on the basis of resistance and exhibit little or no drift in resistance over time. The amorphous or partially amorphous programming state has a much higher resistance than any of the crystalline programming states and may exhibit a drift in resistance over time. The resistance drift of the amorphous or partially amorphous state does not, however, lead to ambiguity in identifying programming states. In one embodiment, drift causes the resistance of the amorphous or partially amorphous state to increase in time at a rate that increases the resistance contrast between the amorphous or partially amorphous state and any of the crystalline programming states. As a result, overlapping of states is avoided.
In one embodiment, the programming method utilizes at least two crystalline programming states and one amorphous programming state. In another embodiment, the programming method utilizes at least two crystalline programming states and a mixed crystalline-amorphous programming state. In still another embodiment, the programming method utilizes at least three crystalline programming states. Additional crystalline, mixed crystalline-amorphous, and amorphous states may be included in the method along with the at least two crystalline states to obtain additional programming states. In one embodiment, the method includes four or more programming states. In another embodiment, the method includes six or more programming states. In a further embodiment, the method includes eight or more programming states.
The crystalline programming states may be different crystalline phases of a phase-change material. In one embodiment, the at least two crystalline programming states include one state that is predominantly face-centered cubic and a second state that is predominantly hexagonal, where the resistances of the face-centered cubic and hexagonal states are distinguishable. In another embodiment, the at least two crystalline programming states include one state that is predominantly face-centered cubic and one state that is a mixture of face-centered cubic and hexagonal crystalline phases. In still another embodiment, the at least two crystalline programming states include one state that is predominantly hexagonal and one state that is a mixture of face-centered cubic and hexagonal crystalline phases. In a further embodiment, the at least two crystalline programming states include one state that is predominantly face-centered cubic, one state that is a mixture of face-centered cubic and hexagonal crystalline phases, and one state that is predominantly hexagonal. Each of the foregoing may further include one or more amorphous or partially amorphous programming states.
The following examples present results showing the presence of multiple crystalline phases for a representative phase-change composition. The examples demonstrate that the crystalline phases are sufficiently resolved in resistance and possess sufficiently low resistance drift to be useful as programming states in a multilevel programming scheme.
In this example, fabrication of the memory device used to obtain the experimental results described hereinbelow is described. The device structure was a standard two-terminal design that included Ge32Sb14Te54 as the active material. Ge32Sb14Te54 is a variable resistance composition drawn from the phase-change class of materials. As explained in the following examples, Ge32Sb14Te54 can be stabilized in multiple crystalline programming states through electrical programming. The multiple crystalline programming states include a face-centered cubic phase, a hexagonal phase, and a series of crystalline phases that include a mixture of the face-centered cubic and hexagonal phases in varying proportions. Ge32Sb14Te54 can also be stabilized in an amorphous programming state and a plurality of mixed crystalline-amorphous programming states.
The device structure is schematically illustrated in
In this example, the I-V (current vs. voltage) and R-I (resistance vs. current) characteristics of the device described in Example 1 are summarized. The I-V and R-I measurements were performed by applying a series of voltage pulses of increasing amplitude between upper electrode 130 and lower electrode 110. The initial voltage pulse had an amplitude of ˜0.1 V and each successive pulse was increased in amplitude by ˜0.1-0.2V up to a maximum voltage of ˜7V. The voltage pulses had a duration of 500 ns. A fixed resistor of ˜2 Kohms was placed in series with the pulse generator to serve as a current limiter. While each voltage pulse was applied, the current passing through the device was measured. After each voltage pulse was applied, the resistance of the device was measured. In this way, a current and resistance was obtained for each voltage pulse. The points were plotted to obtain the I-V relationship shown in
The initial pass of operation was a forming cycle that began with a device in an as-fabricated state. The forming cycle exhibited no switching event in the I-V curve and exhibited an initial resistance of ˜20 kΩ in the R-I plot. After the forming cycle, the I-V data of
The results of this example show that the composition Ge32Sb14Te54 exhibits electrical characteristics consistent with phase-change memory operation. The following examples consider the range and characteristics of programming states available from the device.
This example illustrates the effect of programming conditions on the resistance of the device described in Examples 1 and 2 hereinabove. More specifically, this example demonstrates that the characteristics of the electrical programming pulses can be varied to consistently stabilize a plurality of crystalline programming states. A programming experiment was completed in which the thermal environment of the phase-change material was varied through modification of the quench rate of the phase-change material.
The programming technique involved measurement of the device resistance after application of a two-pulse sequence. The first of the two pulses was a reset pulse having a voltage amplitude sufficient to melt the active phase-change material and a falltime that was sufficiently short to stabilize the phase-change device in its reset state. As noted hereinabove, the reset state is an amorphous programming state that lacks appreciable crystalline content. By decreasing the falltime of the pulse, the phase-change material was cooled at a rate sufficient to quench the material and prevent appreciable crystallization. The device was then subjected to a programming pulse having a fast risetime (steep leading edge) and the same voltage amplitude as the reset pulse, but a slower falltime (gradual trailing edge). After application of the programming pulse, the resistance of the device was measured and recorded. The experiment was repeated for a series of two-pulse sequences in which the falltime of the programming pulse was varied.
The two-pulse programming sequences of this example were designed to induce crystallization and stabilize crystalline phases of the phase-change material. As noted hereinabove, crystallization requires the active material to exist at a temperature between the crystallization temperature and melting temperature for a sufficiently long period of time to permit the structural rearrangements necessary to form an ordered structure from the melt phase. The two-pulse sequences of this example use the falltime of the programming pulse to regulate the rate of cooling of the active material. The amplitude of each programming pulse is sufficient to melt the active material. Programming pulses with a long falltime promote crystallization by insuring that the device exists at a temperature that permits crystallization for a sufficient time. Programming pulses with a sufficiently short falltime cause the active material to cool quickly and prevent it from existing at a temperature that permits crystallization for a sufficient time. Instead, the active material quenches to a state having a substantial amorphous content.
The objective of the programming scheme of this example is to cycle a device over a series of two-pulse programming sequences that use a common initial reset pulse and a second programming pulse whose falltime varies over a range of times. The instant inventors recognize that the time required for crystallization may vary for different crystalline phases and that the initial crystalline phase that may form during programming may be a metastable phase. By considering programming pulses with extended falltimes, the instant inventors have discovered that phase-change compositions can be stabilized in multiple crystalline phases and that programming schemes based on new crystalline phases can be devised.
The sequence of programming pulses used in this example is depicted more specifically in
The ˜60 kΩ device resistance observed after applying a programming pulse with a 100 ns falltime is much lower than the reset resistance and is consistent with a transformation of the device to a crystalline programming state. A 100 ns falltime is much longer than the 3 ns falltime used for the reset pulse and, based on the pronounced reduction in resistance observed in the data, is sufficient to induce crystallization. It is believed by the instant inventors that application of the programming pulse with a 100 ns falltime stabilizes the device in a crystalline programming state in which the phase-change material is in the face-centered cubic crystalline phase identified in the co-pending parent patent application Ser. No. 11/032,345. It is further believed that the crystalline phase consists essential of the face-centered cubic phase as the only crystalline phase and that no appreciable amorphous phase content is present. The 100 ns falltime of the programming pulse is believed to be sufficiently long to permit crystallization of the amorphous phase content and if present, any residual amorphous phase content would have no material effect on the device resistance. The instant inventors note that one could achieve a successful binary memory device by operating the material between the amorphous programming state (˜2.5 MΩ resistance) and the face-centered cubic crystalline programming state (˜60 kΩ resistance).
The data presented in
The results of
In summary, this example shows that multiple crystalline programming states can be stabilized in a phase-change material by controlling programming conditions. Distinct face-centered cubic and hexagonal crystalline phases could be stabilized through appropriate selection of programming conditions. In addition, a continuous range of mixed crystalline phase programming states was observed in which the relative proportions of the face-centered cubic and hexagonal crystalline phases varied continuously.
The data presented in Example 3 indicated that the face-centered cubic crystalline phase, the hexagonal crystalline phase, and a series of mixed crystalline phase programming states could be reproducibly stabilized. The data further indicated that the different crystalline programming states were well-resolved in resistance. In this example, the feasibility of using the crystalline programming states for multilevel operation of a phase-change memory device is demonstrated. As noted hereinabove, overlap of programmed memory states arising from resistance drift currently limits the prospects of multilevel memory devices. This example presents the results of drift measurements performed for the reset state, a mixed amorphous-crystalline programming state, and several crystalline programming states.
The drift experiments were completed by first resetting the device and then programming it to a particular resistance state by applying a programming pulse having a particular falltime. Falltimes ranging from 10 ns to 9 μs were considered. After the device was programmed to a particular resistance state, an initial resistance was measured and the variation of the resistance with time was determined. The data were modeled with the power law formula described hereinabove to obtain a drift coefficient. The drift coefficient was then used to extrapolate the drift data to longer times to assess the feasibility of developing a multilevel programming scheme that includes the crystalline programming states identified in this invention.
To demonstrate multilevel performance free from overlap of states caused by drift, selected programming states from among those included in
The results indicate that the crystalline programming states exhibit low drift coefficients and that the drift in resistance after 10 years is insufficient to cause overlapping of the resistance states. Although drift causes a slight increase in the resistance of the crystalline programming states, the drift for each programming state is insufficient to cause the resistance of the state to increase to the initial resistance observed for the neighboring programming state. As a result, overlapping of states does not occur over an extrapolated time period of at least ten years and a robust multilevel programming scheme is achieved.
The programming scheme demonstrated in this example is based on nine programming states that include a plurality of crystalline programming states. The programming states include an amorphous programming state (e.g. the reset state or the programming state achieved with a programming pulse having a falltime of 10 ns), a mixed crystalline-amorphous state (e.g. the programming state achieved with a programming pulse having a falltime of 50 ns), crystalline programming states that consist essentially of a single crystalline phase (e.g. the programming states achieved with programming pulses having falltimes of 100 ns or 9 μs), and crystalline programming states that include a mixture of crystalline phase (e.g. the programming state achieved with a programming pulse having falltimes of 300 ns, 400 ns, 700 ns, 1 μs, or 2 μs). Other programming schemes including more or less than nine programming states are readily envisionable by those of skill in the art based on the disclosure herein.
The combination of states selected for a programming scheme includes two or more crystalline programming states along with a third programming state and the method of programming includes transforming the phase-change device to each of the programming states at least once. The two or more crystalline programming states may include one or more crystalline phases. In one embodiment, a crystalline programming state consists essentially of one crystalline phase. In another embodiment, a crystalline programming state may include two or more crystalline phases. The programming scheme may include one or more crystalline programming states based on one crystalline phase, one or more crystalline programming states based on a mixture of two or more crystalline phases, or a combination thereof. The third programming state may be a crystalline programming state (based on one crystalline phases or a combination of two or more crystalline phases), an amorphous programming state, or a mixed crystalline-amorphous programming state.
In one embodiment, the magnitude of drift is sufficiently small to prevent overlapping of any two programming states in a multilevel programming scheme for at least one month. In another embodiment, the magnitude of drift is sufficiently small to prevent overlapping of any two programming states in a multilevel programming scheme for at least one year. In still another embodiment, the magnitude of drift is sufficiently small to prevent overlapping of any two programming states in a multilevel programming scheme for at least five years. In a further embodiment, the magnitude of drift is sufficiently small to prevent overlapping of any two programming states in a multilevel programming scheme for at least ten years.
The results described in the foregoing examples indicate that crystalline programming states exhibit low drift and are good candidates for inclusion in multilevel programming schemes. Although the face-centered cubic, hexagonal, and mixtures thereof have been disclosed as illustrative crystalline phases, the instant invention generally extends to electrically-induced transformations among crystalline phases having other crystallographic structures, including stoichiometric and non-stoichiometric compositions as well as crystalline phases including defects or distortions from regular crystallographic structures. Representative crystallographic structures that may be attained by phase-change compositions include orthorhombic, tetragonal, trigonal, triclinic, monoclinic, body-centered cubic, and cubic close packed.
Those skilled in the art will appreciate that the methods and designs described above have additional applications and that the relevant applications are not limited to those specifically recited above. Also, the present invention may be embodied in other specific forms without departing from the essential characteristics as described herein. The embodiments described above are to be considered in all respects as illustrative only and not restrictive in any manner.
This application is a continuation in part of U.S. patent application Ser. No. 11/032,345 entitled “Phase-change Memory that Switches between Crystalline Phases” and filed on Jan. 10, 2005 U.S. Pat. No. 7,923,724, the disclosure of which is incorporated in its entirety herein.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5098761 | Watanabe et al. | Mar 1992 | A |
| 7005665 | Furkay et al. | Feb 2006 | B2 |
| 7923724 | Czubatyj | Apr 2011 | B2 |
| Number | Date | Country | |
|---|---|---|---|
| 20100027328 A1 | Feb 2010 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 11032345 | Jan 2005 | US |
| Child | 12578638 | US |
