Patent Application
20240206186
This application claims foreign priority to European Application No. 22213755.6, filed Dec. 15, 2022, which is incorporated by reference herein in its entirety.
The disclosed technology relates generally to a ferroelectric field-effect transistor (FeFET) memory structure and, more particularly, to a memory device including the FeFET memory structure and a method of fabricating the FeFET memory structure.
A ferroelectric field-effect transistor (FeFET) is a field-effect transistor that incorporates a ferroelectric material as its gate dielectric. The FeFETs can be used in memory devices. For example, data can be stored in the form of a threshold voltage of the FET, which depends on the net ferroelectric polarization (up or down) charge of the ferroelectric material.
The FeFETs can also be used for non-volatile memories (NVM). These memories with the FeFETs can be faster and more energy efficient than conventional flash memories. Some FeFETs can offer faster read/write operations of ca. 10-1000 ns, compared to ca. 100 μs in conventional flash memories, and require lower program/erase voltages, which enables operation with lower energy consumption.
A conventional FeFET is shown in
In view of the above, an objective of the disclosed technology is to provide an improved FeFET memory structure and an improved method of fabricating the FeFET memory structure. In some examples, the disclosed technology can resolve the above-mentioned deficiencies.
These objectives are achieved by the solutions of disclosed technology described in the independent claims. Advantageous implementations are described in the dependent claims.
A first aspect of the disclosed technology is a ferroelectric field-effect transistor (FeFET) memory structure, including a substrate that can include an insulator layer; a gate metal layer on the insulator layer; at least one ferroelectric material layer on the gate metal layer; a layer structure that includes at least one wide bandgap semiconductor layer on the ferroelectric material layer. The layer structure can include: a first section having a first height, and at least one second section having a second height that is smaller than the first height. The FeFET memory structure can further include a drain metal structure which is arranged on the first section of the layer structure; and one or more source metal structures. Each source metal structure can be arranged on a respective second section of the layer structure.
This achieves the advantage that the FeFET memory structure can be provided which requires lower voltages during reading/writing operations and operates with enhanced speed.
In some examples, the ferroelectric material layer of the memory structure can store information by its two net ferroelectric polarization states (up and down). Thereby, the sections of the ferroelectric material layer under each source structure can be individually programmed/read-out. This reading/writing is performed by applying appropriate voltages to the source, the gate and the drain contacts. The ferroelectric polarization state of the ferroelectric material layer (under each source) can determine the local conductive state of the wide bandgap semiconductor layer and its contact resistance to the above source metal structure.
The reduced thickness of the layer structure in the second section below the source metal structure(s) are advantageous. For example, the thinner channel (e.g., the layer structure below the source) can ensure a higher modulation of the conductivity of the channel (depending on the ferroelectric polarization) which causes a higher modulation in the contact resistance with the source contact. Thus, facilitating the read-out of the memory state. Further, the reduced thickness can reduce the programming voltage requirements of the ferroelectric material layer. Apart from the region below the source, the thickness of the layer structure can be increased (e.g., between adjacent sources or under the drain). This enhanced thickness can reduce the sheet resistance of the wide bandgap semiconductor layer for horizontal currents (e.g., currents to and from the drain structure), which can enhance the speed of operation of the memory.
In some examples, the wide bandgap semiconductor layer has a bandgap larger than Si. For instance, the wide bandgap semiconductor layer can be an oxide semiconductor layer (e.g., IGZO). Such oxide based semiconductor materials do not form an additional interfacial layer (e.g., a SiO2 layer). Thus, the programming voltage does not have to be increased to compensate for this interfacial layer. Furthermore, a less favorable band alignment for charge trapping can be achieved. This can prevent the drifting of the threshold voltage and, thus, improves the endurance of the device. Furthermore, the reduced charge trapping can also reduce a delay in reading after writing data operation in the memory.
The source and drain metal structures can include identical or different metals. The drain metal can form an ohmic contact with the first section of the layer structure, while the source metal can form a Schottky contact with the second section of the layer structure. The insulator layer can insulate the memory device.
In some embodiments, the first section of the layer structure has a height between 10 and 30 nm and, in some examples, between 15 and 25 nm. This can achieve the advantage of an enhanced sheet resistance of the wide bandgap semiconductor layer for horizontal currents which can increase the speed of operation of the memory device.
In some embodiments, at least one second section of the layer structure can have a height between 2 and 8 nm and, in some examples, between 3 and 5 nm. In some examples, this reduced layer thickness below the source contacts can cause a strong modulation in the contact resistance between the wide bandgap semiconductor layer and the source metal structures, depending on the net polarization state of the ferroelectric. For example, the difference between the contact resistance between the polarizations “up” and “down” can be enhanced.
In the following, various compounds and materials are disclosed without reference to their stoichiometry. However, it will be understood that such compounds and materials encompass stoichiometric and off-stoichimetry compositions, unless specified otherwise. For example, “PbZrTiO” will be understood to include Pb(ZrxTi1-x)O3, where 0≤x≤1.
In some embodiments, the layer structure can include at least one layer of any one of the following materials: indium gallium zinc oxide (IGZO) indium tin oxide (ITO), indium zinc oxide (IZO), tungsten oxide (WO), zinc oxide (ZnO), indium oxide (InO), tin oxide (SnO), magnesium zinc oxide (MgZnO), Mg- and Al-doped ZnO (MgAlZnO), or gallium zinc oxide (GaZnO).
The layer structure can be a multilayer stack, which can include layers of different materials and conductivities. A bottom layer of the stack (which also covers the area under the source contacts) can have a lower conductivity than the top layer(s). In some examples, the layer stack can be a bilayer stack.
In some embodiments, the ferroelectric material layer can include any one of the following materials: doped or undoped hafnium-zirconium oxide (HZO), for example, lanthanum doped HZO (La:HZO), doped hafnium oxides, complex oxides from the perovskite family, in some examples SrBiTaO (sometimes referred to as “SBT”) , PbZrTiO (sometimes referred to as “PZT”), BiFeO (sometimes referred to as “BFO”), or an alloy or mixture of BiTiO and BaTiO (sometimes referred to as “BTO-BaTiO”), or other ferroelectrics, such as those having a wurtzite structure, such as AlScN or MgZnO.
In some embodiments, the gate metal layer can include any one of the following materials: titanium nitride (TiN), tungsten (W), aluminum (Al), molybdenum (Mo), doped polycrystalline silicon (poly-Si), doped germanium (Ge), platinum (Pt), or ruthenium (Ru).
In some embodiments, the layer structure can include a plurality of second sections which are separated from each other and which are arranged along a line. A respective source metal structure can be arranged on each of the plurality of second sections.
In some embodiments, the FeFET memory structure can further include a gate contact structure arranged on the gate metal layer. The gate contact structure can be configured to apply a voltage to the gate metal layer.
A second aspect of the disclosed technology relates to a memory device, including a plurality of FeFET memory structures according to the first aspect of the disclosed technology and one or more bit line structures on the FeFET memory structures. Each bit line structure can be in contact with one source metal structure of each of the plurality of FeFET memory structures. Each bit line structure can be configured to apply a voltage to the source metal structures with which it is in contact.
The gate metal layer of each FeFET memory structure can form the respective word line structures of the memory device. The gate metal layers (e.g., word lines) and bit line structures can be arranged perpendicular to each other, forming a crossbar array. Thus, the memory device can be a memory array device.
In some examples, the bit line structures can be arranged on the FeFET memory structures.
A third aspect of the disclosed technology is a method of fabricating a ferroelectric field-effect transistor (FeFET) memory structure. The method can include the steps of: forming a gate metal layer on a substrate that the substrate can include an insulating layer; forming at least one ferroelectric material layer on the gate metal layer; forming a layer structure that can include at least one wide bandgap semiconductor layer on the ferroelectric material layer, where the layer structure can include: a first section having a first height, and at least one second section having a second height that is smaller than the first height. The method can further include: forming a drain metal structure which is arranged on the first section of the layer structure; and forming one or more source metal structures. Each source metal structure can be arranged on a respective second section of the layer structure.
In some examples, the method according to the third aspect of the disclosed technology can be used to fabricate the FeFET memory structure according to the first aspect of the disclosed technology.
In some embodiments, the step of forming the layer structure can include: forming the at least one wide bandgap semiconductor layer on the ferroelectric material layer with a uniform height, the uniform height being the first height; and locally reducing the uniform height of the at least one wide bandgap semiconductor layer to the second height to form the at least one second section.
In some embodiments, the first section of the layer structure can have a height between 10 and 30 nm, for example, between 15 and 25 nm.
In some embodiments, at least one second section of the layer structure has a height between 2 and 8 nm, for example, between 3 and 5 nm.
In some embodiments, the gate metal layer can be removed in a region below the drain metal structure prior to forming the at least one ferroelectric material layer.
In some embodiments, the method can further include: forming a gate contact structure on the gate metal layer. The gate contact structure can pass through a gap in the at least one ferroelectric material layer.
The above description with regard to the FeFET memory structure according to the first aspect of the disclosed technology is correspondingly valid for the method according to the third aspect of the disclosed technology.
The above described aspects and implementations are explained in the following description of embodiments with respect to the enclosed drawings.
As described in the above, a conventional FeFET is shown in
Additionally, it can cause a band alignment (e.g., favorable band alignment) that facilitates charge trapping within the ferroelectric layer. This could diminish the memory window's performance, thereby adversely affecting both the endurance and retention characteristics of the device.
The FeFET memory structure 10 can include a substrate 11 which can include an insulator layer 11b. The FeFET memory structure 10 further can include a gate metal layer 13 on the insulator layer 11b, at least one ferroelectric material layer 14 on the gate metal layer 13, and a layer structure 15. The layer structure 15 can also include at least one wide bandgap semiconductor layer. The layer structure 15 can be arranged on the ferroelectric material layer 14 and can include a first section 15-1 having a first height and at least one second section 15-2 having a second height that is smaller than the first height. The FeFET memory structure 10 further can include a drain metal structure 17 which can be arranged on the first section 15-1 of the layer structure 15, and one or more source metal structures 16. In some examples, each source metal structure 16 can be arranged on a respective second section 15-2 of the layer structure 15.
The ferroelectric material layer 14 can be configured to store information by its two net polarization states. Thereby, the sections of the ferroelectric material layer 14 under each source structure 16 can be individually programmed/read-out.
For instance, the ferroelectric material layer 14 may include any one of the following materials: doped or undoped hafnium-zirconium oxide (HZO), in some examples, lanthanum doped HZO (La:HZO), doped hafnium oxides, complex oxides from the perovskite family, such as SrBiTaO (SBT), PbZrTiO (PZT), BiFeO (BFO), or BTO-BaTiO, or other ferroelectrics, such as those having a wurtzite structure, for example, AlScN or MgZnO.
The layer structure 15 on the ferroelectric material layer 14 can include one or more wide bandgap semiconductor layers. In some examples, each wide bandgap semiconductor layer has a bandgap larger than Si. For instance, the wide bandgap semiconductor layers can be oxide semiconductor layers (e.g., IGZO). Such oxide based semiconductor materials do not form an additional interfacial layer (e.g., a SiO2 layer) between the channel and the ferroelectric material layer 14. Thus, the programming voltage may not have to be increased to compensate for this interface layer. Furthermore, a less favorable band alignment for charge trapping can be achieved. This can prevent the drifting of the threshold voltage and thus lead to improved endurance.
In some examples, the memory structure 10 can be a back gated FeFET, where the gate can be located under the source metal contact(s) 16. The layer structure 15 or parts thereof can form a channel of the FeFET memory structure 10.
The wide bandgap semiconductor material layer(s) of the layer structure 15 can include or be formed from any one of the following materials: indium gallium zinc oxide (IGZO) indium tin oxide (ITO), indium zinc oxide (IZO), tungsten oxide (WO), zinc oxide (ZnO), indium oxide (InO), tin oxide (SnO), magnesium zinc oxide (MgZnO), Mg- and Al-doped ZnO (MgAlZnO), or gallium zinc oxide (GaZnO).
In some examples, the polarization (e.g., net polarization) state of the ferroelectric material layer (under each source metal structure 16) determines the local conductive state of the wide bandgap semiconductor layer(s) and its contact resistance to the above source metal structure 16, which can be utilized to read-out the ferroelectric polarization and, thus, memory state.
For instance, the height of the first section 15-1 of the layer structure 15 (e.g., the first height) is between 10 and 30 nm, in some examples, between 15 and 25 nm, and the height of the at least one second section 15-2 of the layer structure 15-2 (e.g., the second height) is between 2 and 8 nm, in some examples, between 3 and 5 nm.
In some examples, the ferroelectric material layer 14 has a bigger thickness than the layer structure 15. In some examples, the ferroelectric material layer 14 has a bigger thickness than the second section 15-2 of the layer structure 15. This can ensure a stronger potential drop across the ferroelectric material layer 15.
In some examples, the reduced thickness of the layer structure 15 below the source metal structure(s) 16 increases the modulation of the conductivity of the channel. The higher conductivity modulation can ensure a higher modulation in the contact resistance between the layer structure 15 and the source metal structure 16 which can be used to measure the polarization state of the ferroelectric. Furthermore, the reduced thickness can reduce programming voltage requirements of the ferroelectric material layer.
The increased thickness of the layer structure 15 in the first section 15-1 can reduce the sheet resistance of the wide bandgap semiconductor layer(s) for horizontal currents (e.g., current to and from the drain metal structure 17).
The insulator layer 11b, in some examples, can insulate the memory structure 10. The insulator layer 11b can be an oxide. The substrate 11 may further include a base substrate 11a, such as a silicon wafer or a part thereof. In some examples, the insulator layer 11b is arranged on the base substrate 11a.
The memory structure 10 may include a further oxide layer 12 which can be arranged on the insulator layer 11b in regions that are not covered by the gate metal layer 13, in some examples under the first section of the layer structure 15. An additional oxide layer 18 or other isolating layer can be arranged above the layer structure 15.
The layer structure 15 can be a single layer which is structured to have the first section 15-1 with the first height and the second section 15-2 with the second height.
In some embodiments, as shown in
For instance, the source metal structure 16 can contact the bottom layer 15a which has a lower conductivity to ensure a higher modulation of its conductivity as a function of the ferroelectric polarization. The further layer 15b on the top which is in contact with the drain metal structure 17 can have a higher conductivity to ensure that the sheet resistance of the further layer 15b, and the contact resistance to the drain metal structure 17 is lower.
The source and drain metal structures 16, 17, respectively, can include identical or different metals. For instance, the drain metal structure 17 can form an ohmic contact with the layer structure 15, while the source metal structure 16 can form a Schottky contact with the layer structure 15
The gate metal layer 13 can include any one of the following materials: titanium nitride (TiN), tungsten (W), aluminum (Al), molybdenum (Mo), doped polycrystalline silicon (poly-Si), doped germanium (Ge), platinum (Pt), or ruthenium (Ru).
In some examples, the gate metal layer forms a word line structure and the source metal structures 16 are connected to a bit line (not shown in
The layer structure 15 may include a plurality of second sections which are separated from each other, and which are arranged along a line. A respective source metal structure 16 can be arranged on each of the plurality of second sections.
Programming of the memory structure 10 can be performed by applying a positive respectively negative bias between the gate and source metal structures 13, 16 (MFSM capacitor). The drain metal structure 17 can be used to pump current through the bit line while reading the memory state.
In some examples, after programming, the net programmed polarization state of the ferroelectric material layer 14 influences both the channel (e.g., the section of the layer structure 15 under the source) as well as the source contact resistance thereby regulating the magnitude of the current flowing through the source metal structure 16 to the bit line. For instance, if the channel is in depletion, the contact resistance between source metal structure 16 and layer structure 15 will be higher and a lower current flows though the source metal structure 16 to the bit line. If the channel is in accumulation, the contact resistance between source metal structure 16 and layer structure 15 will be lower and a higher current flows though the source metal structure 16 to the bit line.
The memory device 40 can include a plurality of FeFET memory structures 10, such as shown in
The memory device 40 can include a cross bar array. The FeFET memory structures 10 can be arranged in vertical columns, and the bit line structures 41 can be arranged in horizontal rows. Thus, the memory device 40 can be a memory array device. Each memory structures 10 (e.g., in the example depicted in
The bit line structures 41 can be connected to one or more sense amplifiers. In some examples, the memory device 40 can be operated with a unipolar power supply.
As shown in
Both channel conductivity and source contact resistance can be modulated by programming the ferroelectric material layer 14, e.g. a HZO layer. In some examples, the ferroelectric material layer 14 is programmed in the regions below the source metal structures 16 (highlighted by the dashed circles in
The proposed design of the memory device 40 can have several advantages over conventional FeFET memories: 1) The regions of the ferroelectric layer (e.g., undoped or doped HZO) under the source metal structures 16 can be selectively programmed by applying a programming voltage directly between the bit lines (via source metal structures 16) and word line (gate metal layer 13); this allows for more efficient programming as compared to programming the entire ferroelectric layer or a longer section thereof. 2) The use of a wide bandgap materials, such as oxides, as channel material can prevent the formation of interfacial oxides which allows for lower programming voltages and causes a lower charge trapping and, thus, higher endurance. 3) Both channel and contact resistance can be modulated. 4) The memory device 40 can be used as a random access memory (RAM). 5) A read disturb can be minimized or prevented by using the separate drain metal contacts 17. 6) The wide bandgap semiconductors used for the layer structure 15 can offer a very large variation in the Schottky barrier height. 7) The memory cell does not have to rely on the availability of the minority carriers. Absence of minority carrier can help to increase the contact resistance ratio between a program and an erase state. 8) The drain metal contact 17 can be common for a string. Thus, area efficiency can be improved.
In addition, a memory cell density can be increased by sequential stacking of layers of memory devices 40. All processing steps to fabricate the memory device 40 can have a low enough thermal budget compatible with BEOL (back end of line) processing.
Each word line WL-1 . . . N can include a word line enable gate EWL-1 . . . N, such as a complementary pass transistor (e.g., NMOS+PMOS). Likewise, each drain line Drain-1 . . . N can include a drain line enable gate ED-1 . . . N, such as a complementary pass transistor (e.g., NMOS+PMOS). The word line enable gates EWL-1 . . . N and the drain line enable gates ED-1 . . . N can selectively activate the respective word lines WL-1 . . . N respectively drain lines Drain-1 . . . N.
Each memory cell 71 of the memory array device 70 can include a source metal structure 16 connected to a word line of the memory device and a section of the gate layer 13 below the source metal structure 16 connected to a bit line. The ferroelectric material layer 14 and the layer structure 15 can be arranged between source metal structure 16 and gate layer 13. Each bit line BL-1 . . . N can be further connected to a sense amplifier.
In some examples, the memory array 70 can be operated in two modes: 1) random access, e.g. for non-volatile DRAM; and 2) block erase, e.g. for storage class NVM.
In some examples, the programming of the string 81 can include a two-step operation: In a first step, as shown in
For example, during writing, all non-selected word lines and all drains are set to Vprog/2, and each bit line potential is set to either 0 V (to write ‘0’) or Vprog (to write ‘1’). To first program a ‘1’ in the center memory cell of the string 81, the selected WL-1 is set to 0 V (cf.,
The read sequence can include the followings: At first, the bit lines BL-1 . . . N are precharged to a precharge voltage Vprecharge which is higher than a reference voltage Vref of the sense amplifier. The selected word line voltage of WL-1 is set to Vread, with |Vread |<|Vprog|. Non-selected WL are set to Vinhibit. The voltage Vinhibit will push the non-selected cells into depletion, thereby suppressing the read disturbance. (|Vinhibit|<|Vprog|). All drain voltages are set to 0 V. In some examples, the non-selected drain lines can be first precharged to 0 V and then be disabled through the pass transistors ED-2 . . . N to reduce interference from non-selected cells. The pass transistor ED-1 of the selected drain line can be enabled. The bit lines BL-1 . . . N will discharge through the selected cells of the string 81. In some examples, the rate of the discharging of the bit line capacitor will be governed by an RC time constant of the contact resistance (Rc) of the selected memory cell and the capacitance of the bit line (CBL).
In some examples, the cells with lower contact resistance (‘0’ programmed) will discharge the bit line capacitance faster compared to the cells with higher contact resistance (‘1’ programmed). The sense amplifier(s) can sample the voltage on the bit lines after a fixed delay. If the voltage on a bit line (Vbitline) is larger than a reference voltage (Vref), then the stored data is ‘1’ else it is ‘0’.
In some examples, depending on the contact resistance in a memory cell, a higher or lower current will flow through the source contact to the bit line.
In some examples, during block erase, all strings of the memory device are selected and a ‘1’ is written into an entire block of memory cells. Therefore, all word lines WL-1 . . . N are set to 0 V, and all bit lines BL-1 . . . N are set to Vprog. The voltage across the gate and the bit line of all memory cells is set to −Vprog (Vgs=−Vprog).
As shown in
The substrate 11 can include an insulating layer 11b. In some examples, the gate metal layer 13 is formed on the insulating layer 11b. The substrate 11 may further include a base layer or base substrate 11a, e.g., a silicon wafer or a part thereof. The insulator layer 11b can be an oxide, e.g. a silicon oxide, and can be deposited on the base layer 11a.
The substrate 11 can further include a FEOL layer 11c. The FEOL layer 11c can include a control circuitry for memory operation, which is, e.g., first fabricated on the base layer 11a. This can allow integration the FeFET memory structure 10 in a BEOL (back end of line) stack as all processes can be compatible to a BEOL thermal budget.
In a second step, shown in
In a further step, shown in
In some examples, the at least one wide bandgap semiconductor layer is formed with a uniform height, the uniform height being the first height
The ferroelectric material layer can be La doped HZO. The wide bandgap semiconductor layer can include an IGZO layer and can form a channel. An optional annealing step in an O2 ambient can be carried out to modulate the conductivity of the channel (in case of an IGZO channel).
The layer structure can then be further processed by locally reducing the uniform height of the at least one wide bandgap semiconductor layer to the second height to form the at least one second section, as shown in
In some embodiments, the layer structure 15 can include multiple wide bandgap semiconductor layers 15a, 15b which are arranged in the form of a layer stack, as schematically depicted in
For instance, the first section of the layer structure 15 has a height between 10 and 30 nm, in some examples between 15 and 25 nm, and the at least one second section of the layer structure 15 has a height between 2 and 8 nm, in some examples between 3 and 5 nm.
In subsequent steps, shown in
Therefore, the windows in the isolating dielectric 18 can be filled by depositing an additional isolation dielectric (cf.,
Subsequently, new windows can be opened in the layer 18 (cf.,
In a further step, shown in
In the claims as well as in the description of disclosed technology, the word “comprising” does not exclude other elements or steps and the indefinite article “a” or “an” does not exclude a plurality. A single element may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation.
| Number | Date | Country | Kind |
|---|---|---|---|
| 22213755.6 | Dec 2022 | EP | regional |
