Thin-film storage transistor with ferroelectric storage layer

Abstract

By harnessing the ferroelectric phases in the charge storage material of thin-film storage transistors of a 3-dimensional array of NOR memory strings, the storage transistors are adapted to operate as ferroelectric field-effect transistors (FeFETs), thereby providing a very high-speed, high-density memory array.

Claims

1. In a 3-dimensional array of memory strings formed above a planar surface of a semiconductor substrate, each memory string comprising: first, second and third transistor material layers, the third transistor material layer being formed to be in contact with both the first and the second transistor material layers; a plurality of conductors; and a ferroelectric storage layer between the conductors and the third transistor material layer, and the third transistor material layer is provided between the ferroelectric storage layer and the first and second transistor material layers, wherein (i) the first, the second and the third transistor material layers, the ferroelectric storage layer and the conductors form a plurality of ferroelectric field-effect transistors (FeFETs) for the memory string; (ii) the first and the second transistor material layers provide a common bit line and a common source line for the FeFETs, respectively; (iii) the third transistor material layer provides a channel region for each FeFET in the memory string, (iv) the ferroelectric storage layer provides a polarizable layer for each FeFET; and (v) each conductor provides a gate electrode for one of the FeFETs in the memory string, wherein each memory string comprises the plurality of FeFETs formed along the common bit line and the common source line, each FeFET being formed at the intersection of each of the conductors to the common bit line and the common source line, and wherein the three dimensional array of memory strings comprises stacks of memory strings formed over the planar surface of the semiconductor substrate, the stacks of memory strings being arrange in a row along a first direction substantially parallel to the planar surface of the semiconductor substrate and each stack being separated from adjacent stacks of memory strings by isolation dielectric layers, the plurality of conductors associated with each memory string being provided in the isolation dielectric layer and extending in a second direction substantially normal to the planar surface of the semiconductor substrate to intersect with the memory strings in each stack, forming the FeFETs at each intersection with the common bit line and the common source line of each memory string in each stack.

2. The memory string of claim 1, wherein the first and second transistor material layers each comprise a semiconductor layer of a first conductivity type, and the third transistor material layer comprises a semiconductor layer of a second conductivity type opposite the first conductivity type.

3. The memory string of claim 1, wherein the first and second transistor material layers each comprise a metal layer, and the third transistor material layer comprises a conductive metal oxide.

4. The memory string of claim 3, wherein the metal layer comprises one or more of titanium nitride-lined tungsten, tungsten, cobalt, and molybdenum.

5. The memory string of claim 3, wherein the conductive metal oxide comprises one or more of: gallium oxides, zinc oxides, and indium oxides.

6. The memory string of claim 5, wherein the indium oxides comprise one or more of: indium gallium zinc oxide (IGZO), indium zinc oxide (IZO) and any conductive metal oxides with charge-carriers mobilities modifiable by inclusion of one or more impurities.

7. The memory string of claim 1, wherein the FeFETs of the memory string are organized as a NOR memory string.

8. The memory string of claim 1, wherein the ferroelectric storage layer comprises an interface dielectric layer and a ferroelectric material layer.

9. The memory string of claim 8, wherein the interface dielectric layer comprises a material with a dielectric constant greater than 3.9.

10. The memory string of claim 8, wherein the interface dielectric layer comprises one or more of zirconium oxide (ZrO.sub.2), silicon oxynitride (SiON), silicon nitride (Si.sub.3N.sub.4), or silicon oxide (SiO.sub.2).

11. The memory string of claim 8, wherein the interface dielectric layer is between 0.0 to 2.0 nm thick.

12. The memory string of claim 8, wherein the interface dielectric layer comprises a native oxide formed inherently when the ferroelectric material layer is directly deposited on the third semiconductor layer.

13. The memory string of claim 8, wherein the interface dielectric layer comprises silicon oxide (SiO.sub.2) and a high-k dielectric material.

14. The memory string of claim 13, wherein the high-k dielectric material comprises zirconium oxide (ZrO.sub.2).

15. The memory string of claim 8, wherein the ferroelectric material layer comprises a zirconium-doped hafnium oxide (HfO.sub.2:Zr; or HZO), an aluminum-doped hafnium oxide (HfO.sub.2:Al), a silicon-doped hafnium oxide (HfO.sub.2:Si) or a lanthanum-doped hafnium oxide (HfO.sub.2:La), or any combination thereof.

16. The memory string of claim 15, wherein the HZO comprises hafnium zirconium oxide (HfZrO), hafnium zirconium oxynitride (HfZrON), hafnium zirconium aluminum oxide (HfZrAlO), any combination thereof, or any other hafnium oxide that includes zirconium impurities.

17. The memory string of claim 8, wherein the 3-dimensional array of memory strings are organized such that the ferroelectric material layer of each FeFET is separated from the ferroelectric material layer of the FeFETs in other memory strings.

18. The memory string of claim 1, wherein the ferroelectric storage layer is deposited on the third semiconductor layer using atomic layer deposition (ALD) techniques at temperature between 200 C. to 330 C.

19. The memory string of claim 18, wherein the temperature is between 270 C. and 330 C.

20. The memory string of claim 18, wherein the ferroelectric storage layer is subject to a post-deposition annealing step at a temperature between 400 C. and 1000 C.

21. The memory string of claim 1, wherein each conductor comprises tungsten (W), molybdenum (Mo), aluminum (Al), ruthenium (Ru), tantalum (Ta), titanium (Ti), or any combination or alloy of thereof.

22. The memory string of claim 1, wherein each FeFET has a conducting state threshold voltage greater than 0.0 volts.

23. The memory string of claim 22, wherein the third transistor material layer is boron-doped.

24. The memory string of claim 1, wherein each FeFET has a 1.0 volts to 2.0 volts window between its threshold voltage in its conducting state and its threshold voltage in its non-conducting state.

25. The memory string of claim 1, wherein the third transistor material layer is floating during a programming operation wherein the programming operation is conducted in conjunction with voltage biases that provide a gate-induced drain leakage (GIDL) effect.

26. The memory string of claim 1, wherein at least a portion of the ferroelectric storage layer is deposited using a selective atomic layer deposition technique involving self-assembled monolayers (SAMs) acting on the isolation dielectric layers.

27. The memory string of claim 1, wherein the SAMs comprise species having hydroxyl terminations.

28. The memory string of claim 1, wherein the FeFETs of the memory string are arranged along the first direction substantially parallel to the planar surface.

29. The memory string of claim 1, wherein the FeFETs of a stack of memory strings are arranged along the second direction substantially normal to the planar surface.

30. A thin-film ferroelectric field-effect transistor (FeFET) in a 3-dimensional memory array comprising: a metal source line and a metal bit line provide above a planar surface of a semiconductor substrate, the metal source line and the metal bit line being arranged along a direction substantially parallel to the planar surface of the semiconductor substrate, the metal source line being separated from the metal bit line by an insulator layer in a direction substantially normal to the planar surface of the semiconductor substrate; a channel region formed out of an oxide semiconductor material, the channel region being arranged along the direction substantially normal to the planar surface of the semiconductor substrate; a gate conductor layer arranged along the direction substantially normal to the planar surface of the semiconductor substrate; and a ferroelectric storage layer provided between the channel region and the gate conductor layer, wherein the channel region is in contact with the metal source line and the metal bit line on a first side of the channel region and is in contact with the ferroelectric storage layer on a second side, opposite the first side, of the channel region, the channel region separating the metal source line and the metal bit line from the ferroelectric storage layer.

31. The thin-film FeFET of claim 30, wherein the oxide semiconductor material comprises indium zinc oxide (InZnO; or IZO).

32. The thin film FeFET of claim 30, wherein the channel region is 8.0 to 12.0 nm thick.

33. The thin-film FeFET of claim 30, wherein the channel region has an electron mobility greater than or equal to 12.0 cm.sup.2/V when the channel region has a thickness greater than 6.0 nm.

34. The thin-film FeFET of claim 30, wherein the metal source line or the metal bit line comprises molybdenum.

35. The thin-film FeFET of claim 30, wherein the FeFET is one of a plurality of FeFETs provided on a memory string.

36. The thin-film FeFET of claim 35, wherein the memory string is organized as a NOR memory string of FeFETs.

37. The thin-film FeFET of claim 35, wherein the memory string is one of a plurality of memory strings formed above the planar surface of a semiconductor substrate.

38. The thin-film FeFET of claim 37, wherein the FeFETs of each memory string are arranged along the direction substantially parallel to the planar surface.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) FIG. 1 is an energy band diagram of a conventional storage transistor, which includes multiple sublayers of dielectric materials and stores charge between a channel region and a gate electrode.

(2) FIG. 2 shows some typical direct tunneling current densities (gate currents) for various silicon dioxide thicknesses under different bias conditions.

(3) FIGS. 3(a) and 3(b) illustrate, respectively, direct tunneling of electrons into charge-trapping sublayer 112 and out of charge-trapping sublayer 112 during programming and erase operations.

(4) FIG. 4 illustrates an evolution of the programming window in a storage transistor over more than 10.sup.9 program and erase cycles, showing program state threshold voltage 401 and erase state threshold voltage 402.

(5) FIG. 5 is an energy band diagram in an exemplary storage transistor having channel region 501, tunnel dielectric layer 502 and charge-trapping layer 503.

(6) FIGS. 6(a), 6(b) and 6(c) show, respectively, (i) the lowest energy levels of the conduction bands at substrate 501, tunnel dielectric 502 and charge-trapping layer 503 of a storage transistor; (ii) the lowest energy levels in the conduction bands of these layers of the storage transistor without application of a voltage; and (iii) the electron energy offset 515 between substrate 501 and charge-trapping layer 503, when an erase voltage is applied.

(7) FIGS. 7(a), 7(b) and 7(c) show, respectively, (i) the relative conduction band offsets at substrate 601, tunnel dielectric 602, low conduction band offset (LCBO) barrier layer 603, and charge-trapping layer 604 of a storage transistor; (ii) an energy band diagram of these layers of the storage transistor without application of a voltage; and (iii) electron energy offset 615 between substrate 601 and charge-trapping layer 604, when an erase voltage is applied.

(8) FIGS. 8(a), 8(b) and (c) illustrate the conduction band offset parameters for dielectric layers 602-604 illustrated in FIGS. 7(a)-7(c).

(9) FIG. 9(a) illustrates direct tunneling and FIGS. 9(b) and 9(c) illustrate MFN tunneling in the storage transistor of FIGS. 7(a)-7(c).

(10) FIGS. 10(a) and 10(b) are band diagrams for the structure during programming and erase operations, based on a one-volt drop across tunnel dielectric layer 602 (i.e., b=1 eV, during a programming operation and b=1 eV, during an erase operation).

(11) FIGS. 11(a), 11(b), 11(c) and 11(d) show various simulation results for storage transistors of the present invention.

(12) FIG. 12(a) is an energy band diagram for the conduction band of a gate stack of a storage transistor during an erase operation.

(13) FIG. 12(b) is an energy band diagram for the conduction band of a gate stack of a storage transistor during an erase operation, the storage transistor having additional aluminum oxide layer 607 in blocking dielectric layer 610, according to one embodiment of the present invention.

(14) FIG. 13(a) shows a cross-section of a 3-dimensional array 1300 of NOR memory strings, which may be formed out of the thin-film storage transistors discussed herein, according to one embodiment of the present invention.

(15) FIG. 13(b) shows active stacks 1351-1 and 1351-2 of NOR memory strings in a 3-dimensional array 1350, with each NOR memory string including numerous FeFETs as storage transistors, according to one embodiment of the present invention.

(16) FIG. 14(a) shows hysteresis in the drain current in response to the gate voltage in a conventional FeFET.

(17) FIG. 14(b) shows the desirable hysteresis in the drain current (Ia) response to an applied gate voltage (V.sub.g) in a thin-film FeFET in a NOR memory array, according to one embodiment of the present invention.

(18) FIGS. 15(a), 15(b), 15(c), and 15(d) illustrate a first process that forms a 3-dimensional memory array of FeFETs organized as NOR memory strings, according to one embodiment of the present invention.

(19) FIGS. 16(a), 16(b), 16(c), and 16(d) illustrate a second process that forms a 3-dimensional memory array of thin film FeFETs organized as horizontal NOR memory strings, according to one embodiment of the present invention.

(20) FIGS. 17(a), 17(b), 17(c), 17(d), 17(e), 17(f), and 17(g) illustrate a third process that forms a 3-dimensional memory array of FeFETs organized as NOR memory strings, according to one embodiment of the present invention.

(21) To facilitate cross-reference among the figures, like elements are assigned like reference numerals.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

(22) The above-referenced Related Application discloses 3-dimensional arrays of NOR memory strings, each being formed out of thin-film storage transistors. Provisional Application III further discloses, for example, various methods for fabricating such 3-dimensional arrays of NOR memory strings. These 3-dimensional arrays may be formed, for example, above a planar surface of a semiconductor substrate. In this detailed description, a Cartesian coordinate system is adopted to facilitate clarity in the spatial relationships among features shown in the figures. In this coordinate system, the Z-direction corresponds to a direction substantially normal to the planar surface, and the X- and Y-directions correspond to directions orthogonal to each other and to the Z-direction. The storage transistors of such NOR memory strings may be programmed and erased in 100 nanosecond (ns) or less, rendering them suitable for use in many applications of conventional volatile memory devices, such as dynamic random-access memory (DRAM) devices. These thin-film storage transistors of the Related Application also have an advantage of a retention time in minutes, rather than milliseconds, as typical of conventional DRAM devices. Therefore, one may refer to these thin-film storage transistors as quasi-volatile storage transistors. In many applications, such a quasi-volatile storage transistor should preferably have a high endurance (e.g., in the range of 10.sup.11 cycles) and may preferably be programed or erased using voltages of about 8-9 volts or lower.

(23) Fast programming and fast erase operations require relatively high currents through the gate stack of the storage transistor. FIG. 1 is an energy band diagram of a portion of a storage transistor, which includes multiple sublayers of dielectric materials and stores charge between a channel region and a gate electrode. As shown in FIG. 1, various materials 120 between channel region 110 and gate electrode 114 allow data storage in the storage transistor. These materials include tunnel dielectric sublayer 111, charge-trapping sublayer 112 (e.g., silicon nitride) and blocking dielectric sublayer 113 (e.g., silicon oxide). Charge-trapping sublayer 112 and blocking dielectric sublayer 113 may each be, for example, 4 nm thick. In FIG. 1, line 101 traces the lowest energy states in the conduction bands of the various materials and line 102 traces the highest energy states in the valence bands of the various materials. In such a system, to change the threshold voltage of the storage transistor by one volt in 100 ns requires a programming current density of about 5.0 amps/cm.sup.2. Using silicon dioxide as tunneling dielectric sublayer 111, such a high current density may be achieved at a moderate electric field in the range of 10.0 MV/cm through a direct tunneling mechanism.

(24) FIG. 2 shows some typical direct tunneling current densities (gate currents) for various silicon dioxide thicknesses under different bias conditions. As shown in FIG. 2, the desired high current density (e.g., 5.0 amps/cm.sup.2) may be achieved in silicon dioxide for thicknesses less than 1.5 nm, even for voltages below 1.5 volts across the silicon dioxide layer.

(25) FIGS. 3(a) and 3(b) illustrate, respectively, direct tunneling of electrons into charge-trapping sublayer 112 and out of charge-trapping sublayer 112 during programming and erase operations. As illustrated in FIG. 3(a), the applied programming voltage across gate electrode 114 and channel region 110 lowers the conduction bands of tunnel dielectric sublayer 111, charge-trapping sublayer 112 and blocking dielectric sublayer 113 relative to channel region 110. Specifically, the lowest energy levels in the conduction band of charge-trapping sublayer 112 is lowered to slightly below that of channel region 110 to allow electrons with energy at the lowest energy levels of the conduction band in channel region 110 to direct-tunnel into charge-trapping sublayer 112, as illustrated by arrow 301 in FIG. 3(a).

(26) Likewise, as illustrated in FIG. 3(b), the applied erase voltage across gate electrode 114 and channel region 110 raises the lowest energy levels in the conduction bands of tunnel dielectric sublayer 111, charge-trapping sublayer 112 and blocking dielectric sublayer 113 relative to those of channel region 110. The electric field imparts energy to electrons at the allowed energy levels of the charge-trapping sites in charge-trapping sublayer 112 to direct tunnel into channel region 110, as illustrated by arrow 302 in FIG. 3(b).

(27) The direct tunneling mechanisms by electrons illustrated in FIGS. 3(a) and 3(b) allow fast programming and erase. In contrast, erasing by holes is a slow mechanism. In a floating-substrate quasi-volatile storage cell (e.g., the thin-film storage transistor disclosed in the Related Application), for example, holes in channel region 110 are insufficient to provide an adequate hole current into charge-trapping sublayer 112; the likely erase mechanism for such a storage transistor pulls electrons out from charge-trapping sublayer 112.

(28) In a storage transistor, the voltage difference between the threshold voltages of the storage transistor in the erased state and in the programmed state is referred to as the programming window. The programming window narrows or closes with the number of cycles the storage transistor has been programmed and erased. Such programming window narrowing is due to, for example, degradation at the interface between channel region 110 and tunnel dielectric 111, as a result of interface states formation. Programming window narrowing may also result from charge-trapping at other material interfaces, e.g., between charge-trapping sublayer 112 and blocking dielectric sublayer 113. The endurance of a storage transistor refers to the number of program-erase cycles before the storage transistor fails to maintain an acceptable programming window. As illustrated in FIG. 3(a), electrons that direct tunnel from channel region 110 to charge-trapping sublayer 112 have low energy entering charge-trapping sublayer 112, so that they give up only a small amount of energy settling in the lowest allowed energy states in charge-trapping sublayer 112 (i.e., the lowest energy levels in the conduction bands of channel region 110 and charge-trapping sublayer 112 are very close in the presence of the programming voltage). This energy loss does not cause any appreciable damage to charge-trapping sublayer 112. In contrast, as illustrated in FIG. 3(b), the energy loss by electrons entering channel region 110 during an erase operation is significantly larger. The greater energy loss generates energetic holes (hot holes) in channel region 110, which would be driven by the electric field of the erase voltage towards gate electrode 114. These hot holes create interface traps at the interface between channel region 110 and tunnel dielectric sublayer 111. These interface traps are detrimental to the endurance of the storage transistor and, in fact, may be primarily responsible for the programming window closing. One of ordinary skill in the art may also recognize that the hot holes phenomenon, known as the anode hot hole injection mechanism, provides one model for dielectric breakdown.

(29) FIG. 4 illustrates the evolution of the programming window in a storage transistor over more than 10.sup.9 program and erase cycles, showing program state threshold voltage 401 and erase state threshold voltage 402.

(30) The present invention improves endurance in a storage transistor to exceed 10.sup.11 program-erase cycles using a device structure that ensures electrons tunneling out of a charge-trapping layer into the channel region of the storage transistor (e.g., during an erase operation) are within a desirable low energy range (cool electrons), such that any resulting hole generations are also low-energy and are thus less damaging to the programming window. The device structure provides a substantial direct tunneling programming current density exceeding 1.0 amps/cm.sup.2 (e.g., 5.0 amps/cm.sup.2). The present invention is particularly advantageous for use in storage layers of thin-film storage transistors that are formed in 3-dimensional memory structures, such as those quasi-volatile storage transistors in the 3-dimensional arrays of NOR memory strings disclosed in the Related Application discussed above.

(31) One embodiment of the present invention is illustrated by the model of FIG. 5, showing the conduction and valence energy band boundaries 511 and 512 of an exemplary storage transistor having channel region 501, tunnel dielectric layer 502 and charge-trapping layer 503. As FIG. 5 illustrates, arrow 514 represents electrons direct tunnel from charge-trapping layer 503 to channel region 501. The energy difference (conduction band offset) between the lowest energy levels in the conduction bands of charge-trapping layer 503 and channel region 501indicated by reference numeral 515is the expected energy loss by an electron so tunneled.

(32) The present invention may be achieved by judiciously selecting a combination of materials for a tunnel dielectric material and a charge-trapping dielectric material, to obtain desirable conduction band offsets at these layers relative to the semiconductor substrate (i.e., the channel region) of the storage transistor. FIG. 6(a) shows the lowest energy levels of the conduction bands at substrate 501, tunnel dielectric 502 and charge-trapping layer 503 of the storage transistor. FIG. 6(b) shows the lowest energy levels in the conduction bands of these layers of the storage transistor without application of a voltage. FIG. 6(c) shows the electron energy offset 515 between substrate 501 and charge-trapping layer 503, when an erase voltage is applied. Electron energy offset 515 depends on conduction band offsets between substrate 501 and each of tunnel dielectric layer 502 and charge-trapping layer 503, as well as on the voltage applied for the erase operation. As illustrated in FIG. 6(c), for tunnel dielectric layer 502, using different charge-trapping materials as charge-trapping layer 503, with different conduction band offsets relative to the substrate layer 501, results in greater or lesser energy loss in the tunneling electrons reaching substrate 501. Likewise, for charge-trapping layer 503, using different tunnel dielectric materials as tunnel dielectric layer 502, with different conduction band offsets relative to the substrate layer 501, also results in greater or lesser energy loss in the tunneling electrons reaching substrate 501.

(33) Tunnel dielectric layer 502 may be as thin as 5-40 and may be formed out of silicon oxide (e.g., SiO.sub.2), silicon nitride (SiN), or silicon oxynitride (SiON). A silicon oxide tunnel dielectric layer may be formed using conventional oxidation techniques (e.g., a high-temperature oxidation), chemical synthesis (e.g., atomic layer deposition (ALD)), or any suitable combination of these techniques. A reactive O.sub.2 process may include an ozone step (e.g., using pulsed ozone) for a precisely controlled thickness and an improved oxide quality (e.g., reduced leakage due to defect sites). The ozone step augments solidification of the oxide in a conformal manner, which is particularly advantageous for three-dimensional transistor structures. An annealing step (e.g., an H.sub.2 anneal, a NH.sub.3 anneal, or a rapid thermal annealing) may also fortify tunnel dielectric layer 502. A silicon nitride tunnel dielectric layer may be formed using conventional nitridation, direct synthesis, chemical synthesis (e.g., by ALD), or any suitable combination of these techniques. A plasma process may be used for a precisely controlled thickness and an improved dielectric quality (e.g., reduced leakage due to defect sites).

(34) Tunnel dielectric layer 502 may also include an additional thin aluminum oxide (Al.sub.2O.sub.3) layer (e.g., 10 or less). This additional aluminum oxide layer in the tunnel dielectric layer may be synthesized in the amorphous phase, to reduce leakage due to defect sites.

(35) The following materials may be used to provide tunnel dielectric layer 502 and charge-trapping layer 503:

(36) TABLE-US-00001 Material Conduction Band Offset Silicon oxide (SiO.sub.2) 3.15 eV Hafnium oxide (HfO.sub.2) 1.5 eV Silicon Nitride (Si.sub.3N.sub.4) 2.4 eV Yttrium oxide (Y.sub.2O.sub.3) 2.3 eV Zirconium oxide(ZrO.sub.2) 1.4 eV Zirconium silicon oxide (ZrSiO.sub.4) 1;5 eV Lanthanum oxide (La.sub.2O.sub.3) 2.3 eV Silicon oxinitrides (SiN:H) 1.3-2.4 eV Tantalum oxide (Ta.sub.2O.sub.5) 0.3 eV Cerium oxide (CeO.sub.2) 0.6 eV Titanium oxide (TiO.sub.2) 0.0 eV Strontium titanium oxide (SrTiO.sub.3) 0.0 eV Silicon-rich silicon nitride (SiN:Si) 1.35 eV Silicon nanodots 0.0 eV Ruthenium nanodots 0.7 eV Cobalt nanodots 1.0 eV

(37) Using a lower conduction band offset in the charge-trapping layer provides an effective increase in tunneling barrier in the tunnel dielectric layer, resulting in improved data retention.

(38) Alternatively, a barrier material of low conduction band offset may be introduced into the storage transistor between the tunnel dielectric layer and the charge-trapping layer. FIGS. 7(a)-7(c) are band diagrams representative of such a structure. FIG. 7(a) shows the relative conduction band offsets at substrate 601, tunnel dielectric 602, low conduction band offset barrier dielectric 603 and charge-trapping layer 604 of the storage transistor. FIG. 7(b) is an energy band diagram of these layers of the storage transistor without application of a voltage. FIG. 7(c) shows the electron energy offset 615 between substrate 601 and charge-trapping layer 604, when an erase voltage is applied. Electron energy offset 615 depends on conduction band offsets between substrate 601 and each of tunnel dielectric layer 602, low conduction band offset barrier layer 603 and charge-trapping layer 604, as well as on the voltage applied for the erase operation. As shown in FIGS. 7(a)-7(c), low conduction band offset (LCBO) barrier dielectric 603 preferably has a conduction band offset relative to substrate 601 that is lower than those of both tunnel dielectric layer 602 and charge-trapping layer 604. Judiciously choosing the materials for tunnel dielectric layer 602, LCBO barrier layer 603, and trapping layer 604, cool electron direct tunneling may be achieved for both program and erase operations, resulting in a high endurance in the storage transistor.

(39) FIGS. 8(a)-(c) illustrate the conduction band offset parameters for dielectric layers 602-604 illustrated in FIGS. 7(a)-7(c). As shown in FIG. 8(a), (i) parameter B represents the conduction band offset of tunnel dielectric layer 602 relative to substrate 601, (ii) parameter a represents the conduction band offset of LCBO barrier layer 603 relative to the conduction band offset of tunnel dielectric layer 602, (iii) parameter d represents the conduction band offset of LCBO barrier layer 603 relative to substrate 601, and (iv) parameter c represents the conduction band offset of charge-trapping layer 604 relative to substrate 601. According to one embodiment of the present invention, the conduction band offset of LCBO barrier layer 603 should not be greater than the conduction band offset of charge-trapping layer 604 (i.e., dc) to allow a substantial direct tunneling programming current density exceeding 1.0 amps/cm.sup.2 (e.g., 5.0 amps/cm.sup.2).

(40) FIG. 8(b) shows sloping of the energy level at the bottom of the conduction band of tunnel dielectric layer 602 as a result of the programming voltage. The sloping lowers the energy level of tunnel dielectric layer 602 by parameter b over the thickness of tunnel dielectric layer 602. For the programming operation to be effectuated by direct tunneling, the value of parameter b should be greater or equal to the value of parameter c (i.e., bc). The value of parameter b (in eV units) is the product of the voltage drop across tunnel dielectric layer 602 and the electron charge q (i.e., 1.610.sup.19 coulombs).

(41) When the voltage drop across tunnel dielectric 602 is less than the conduction band offset of charge-trapping layer 604 (i.e., b<c), the tunneling barrier becomes wider, as at least a part of LCBO barrier layer 603 remains a tunneling barrier. In that case, direct tunneling may give way to a modified Fowler-Nordheim (MFN) mechanism, which provides a much smaller current than direct tunneling (e.g., less than 0.1 amps/cm.sup.2).

(42) FIG. 9(a) illustrates direct tunneling, under application of a programming voltage, and FIGS. 9(b) and 9(c) illustrate MFN tunneling, under a lower voltage (intermediate) and an even lower voltage, respectively, in the storage transistor of FIGS. 7(a)-7(c). One may recognize that MFN tunneling may occur in a region of low voltage disturbs during operations of the storage transistor. However, for a storage transistor having the structure illustrated in FIGS. 7(a)-7(c), this MFN tunneling current can be very low for a range of voltages applied. The materials and the thicknesses for charge-trapping layer 604 and barrier layer 603 are selected such that read disturb voltages, programming inhibit voltages or erase inhibit voltages fall within the range of low or intermediate voltages that restrict tunneling to the MFN mechanism.

(43) Thus, the storage transistor of the present invention provides an important advantage: high currents at the programming voltage due to direct tunneling, while having merely a low MFN tunneling current when exposed to a low voltage. This characteristic reduces disturbs during read, programming inhibit, or erase inhibit operations and improves data retention and endurance, particularly in quasi-volatile storage transistors of the present invention that use direct tunneling for fast programming and fast erase operations. In this regard, LCBO barrier layer 603 improves endurance by enabling cool electron-erase operations, which reduces device degradation, as the resulting holes generated in the channel region are low-energy.

(44) By restricting tunneling at low voltages to MFN tunneling, LCBO barrier layer 603 also improve data retention and reduces read disturb, programming-inhibit disturbs and erase-inhibit disturbs, as the read disturbs, programming-inhibit disturbs and erase-inhibit disturbs all occur at low voltages. For example, programming-inhibit disturbs and erase-inhibit disturbs occur at half-select or a lower voltage than that used in the respective programming and erase operations. All these benefits accrue in the storage transistors biased at low voltages, while at the same time maintaining the advantages of the high efficiency of direct tunneling accrue in the storage transistors biased at the higher read, programming or erase voltages.

(45) FIG. 8(c) shows sloping of the energy level at the bottom of the conduction band of tunnel dielectric layer 602 during an erase operation. The sloping raises the energy level of tunnel dielectric layer 602 by parameter b over the thickness of tunnel dielectric layer 602. During the erase operation, electrons in direct tunneling from charge-trapping layer 604 to substrate 601 loses an energy represented by parameter A, which is given by: A=b+c. Note that the conduction band offset of charge-trapping layer 604 should be greater than the amount by which the energy level of a charge-trapping site is below the conduction band of charge-trapping layer 604 in order for the electrons at the charge-trapping site to be included in the direct tunneling current.

(46) According to one embodiment of the present invention, substrate 601 may be implemented by a P-doped silicon, tunnel dielectric layer 602 may be implemented by a 1-nm thick SiO.sub.2 layer (B=3.15 eV), low conduction band offset barrier layer 603 may be implemented by a 2-nm thick Ta.sub.2O.sub.5 layer (d=0.3 eV), charge-trapping layer 604 may be implemented by a 4-nm thick silicon-rich silicon nitride (i.e., SiN:Si; c=1.35 eV), and another 4-nm thick SiO.sub.2 layer may be used to provide a blocking dielectric layer. Unlike silicon nitride (stoichiometrically, Si.sub.3N.sub.4), silicon-rich silicon nitride includes silicon as impurity, which reduces silicon nitride's band gap from 4.6 eV to about 3.6 eV for silicon-rich silicon nitride. Also, silicon nitride has a refractive index of 2.0, while silicon-rich silicon nitride has a refractive index in the range of 2.1-2.3. Gate electrode 606 may be implemented by a highly-doped P-type polysilicon. FIGS. 10(a) and 10(b) are band diagrams for the structure during programming and erase operations, based on a one-volt drop across tunnel dielectric layer 602 (i.e., b=1 eV, during a programming operation and b=1 eV, during an erase operation). As shown in FIG. 10(b), as indicated by arrow 1001, an electron reaching substrate 601 by direct tunneling loses about 1.4 eV of energy during the erase operation. Scattering in LCBO barrier layer 603, as indicated by arrow 1002, may further reduce this energy loss.

(47) According to another embodiment of the present invention, substrate 601 may be implemented by a P-doped silicon, tunnel dielectric layer 602 may be implemented by a 1-nm thick SiO.sub.2 layer (B=3.15 eV), low conduction band offset barrier layer 603 may be implemented by a 2-nm thick CeO.sub.2 layer (d=0.6 eV), charge-trapping layer 604 may be implemented by a 4-nm thick silicon-rich silicon nitride (i.e., Si.sub.3N.sub.r4:Si; c=1.35 eV), and another 5-nm thick SiO.sub.2 layer may be used to provide a blocking dielectric layer 605. Gate electrode 606 may be implemented by a highly-doped P-type polysilicon.

(48) FIG. 11(a)-11(d) show various simulation results for storage transistors of the present invention.

(49) FIG. 11(a) shows a simulation of a storage transistor having a 0.8 nm thick silicon oxide tunneling dielectric layer, a 2.0 nm thick zirconium oxide LCBO barrier layer and a 5.0 nm thick silicon-rich silicon nitride trapping layer. FIG. 11(a) shows that a direct-tunneling current density exceeding 1.0 amps/cm.sup.2 is achieved with a programming voltage around 3.1 volts.

(50) FIG. 11(b) shows a simulation of a storage transistor having a 1.0 nm thick silicon oxide tunneling dielectric layer, a 2.0 nm thick cerium oxide LCBO barrier layer and a 4.0 nm thick silicon-rich silicon nitride trapping layer. FIG. 11(b) shows that a direct-tunneling current density exceeding 1.0 amps/cm.sup.2 is achieved with a programming voltage around 1.6 volts.

(51) FIG. 11(c) shows a simulation of a storage transistor having a 1.0 nm thick silicon oxide tunneling dielectric layer, a 2.0 nm thick tantalum oxide LCBO barrier layer and a 4.0 nm thick silicon-rich silicon nitride trapping layer. FIG. 11(c) shows that a direct-tunneling current density exceeding 1.0 amps/cm.sup.2 is achieved with a programming voltage around 1.8 volts.

(52) FIG. 11(d) shows a simulation of a storage transistor having a 1.0 nm thick silicon nitride tunneling dielectric layer, a 2.0 nm thick cerium oxide LCBO barrier layer and a 4.0 nm thick silicon-rich silicon nitride trapping layer. FIG. 11(d) shows that a direct-tunneling current density exceeding 1.0 amps/cm.sup.2 is achieved with a programming voltage around 2.1 volts.

(53) FIG. 12(a) illustrates a reverse injection electrons phenomenon that may occur during an erase operation. The reverse injection electrons may affect endurance adversely. FIG. 12(a) is an energy band diagram for the conduction band of a gate stack in a storage transistor during an erase operation. As shown in FIG. 12(a), the gate stack includes substrate 601, tunnel dielectric 602, LCBO barrier dielectric 603, charge-trapping layer 604, blocking dielectric layer 605 and gate electrode 606. (Blocking dielectric layer 605 may be, for example, silicon oxide (SiO.sub.2)). During an erase operation, the relatively high electric field across blocking dielectric layer 605 may cause high-energy electronsindicated in FIG. 12(a) by arrow 1201to tunnel from the gate electrode into charge-trapping layer 604, or even into tunnel dielectric layer 602. These reverse injection electrons may damage these layers, adversely affecting the storage transistor's endurance.

(54) According to one embodiment of the present invention, reverse injection electrons may be significantly reduced or substantially eliminating by including a layer of material with a high dielectric constant (high-k material), such as aluminum oxide (Al.sub.2O.sub.3) in the blocking dielectric layer (e.g., blocking dielectric layer 605 of FIG. 10(a)). In that embodiment, a high work function metal (e.g., greater than 3.8 eV, preferably not less than 4.0 eV) may be used for gate electrode. A high-k material of t.sub.H provides an equivalent oxide thickness t.sub.EOT given by:

(55) $t_{EOT}=t_H\frac{{}_{ox}}{{}_H}$
where .sub.ox and .sub.H are the relative dielectric constants of silicon oxide and the high-k material, respectively. Thus, a high-k material can provide the same desirable transistor characteristics (e.g., gate capacitance) at a thickness of t.sub.H, without incurring undesirable leakage of its silicon oxide layer counterpart at the much thinner equivalent thickness t.sub.EOT.

(56) FIG. 12(b) is an energy band diagram for the conduction band of a gate stack in a storage transistor during an erase operation, the storage transistor having additional aluminum oxide layer 607 in blocking dielectric layer 610, according to one embodiment of the present invention. In FIG. 12(b), blocking dielectric layer 610 includes aluminum oxide layer 607 and silicon oxide layer 608. In one implementation, blocking dielectric layer 610 has an equivalent oxide thickness that is substantially the same as blocking layer dielectric 605 of FIG. 12(a). However, as aluminum oxide has a relative dielectric constant of 9.0, while silicon oxide's relative dielectric constant is 3.9, the actual combined physical thickness of aluminum oxide 607 and silicon oxide 608 in FIG. 12(b) is greater than the thickness of blocking dielectric layer 605 of FIG. 12(a). Because high-k dielectric layer 607 has a greater relative dielectric constant than silicon oxide layer 608, the electric field is lower in high-k dielectric layer 607 than in silicon oxide layer 608. The greater combined physical thickness of blocking dielectric layer 610 of FIG. 12(b)which provides a wider tunneling barrier between gate electrode 606 and charge-trapping layer 604and a lower electric field at the interface between gate electrode 606 and high-k material 607 reduce or eliminate reverse injection electrons, thereby resulting in an improved endurance. With high-k electric layer 607 (e.g., aluminum oxide), a high work function metal is preferred for gate electrode 606. The high work function metal creates a high barrier (indicated by barrier height 1202 in FIG. 12(b)) at the gate electrode-aluminum oxide interface, which significantly reduces reverse electron injection the erase operation. Suitable high work function metals include: tungsten (W), tantalum nitride (TaN), tantalum silicon nitride (TaSiN).

(57) FIG. 13(a) shows a cross-section of a 3-dimensional array 1300 of NOR memory strings, which may be formed out of the thin-film storage transistors discussed above, according to one embodiment of the present invention. As shown in FIG. 13(a), stacks 1301-1 and 1301-2 of NOR memory strings are formed above a planar surface of silicon substrate 1302. Stacks 1301-1 and 1301-2 represent a row of any suitable number of active stacks (e.g., 2, 4, 8, 16 . . . ) separated from each other by isolation dielectric (e.g., silicon oxycarbide (SiOC)) layers 203 along the X-direction. Each of the active stacks may include any suitable number (e.g., 2, 4, 8, 16, . . . ) active multi-layers, each providing any suitable number of storage transistors (e.g., 8, 16, . . . , 2048, 4096, . . . )organized as one or more NOR memory stringsseparated from each other along the Y-direction. For example, stack 1301-1 is shown in FIG. 13(a) to include NOR memory strings 204-1 to 204-4. The inset of FIG. 13(a) shows a cross-section of storage transistor 1303 in one NOR memory string of active stack 1301-2.

(58) As shown in FIG. 13(a), storage transistor 1303 includes (i) conductor layer 204a (e.g., a titanium nitride-lined tungsten layer), (ii) N.sup.+-doped amorphous silicon or polysilicon layer 204b (e.g., phosphorus or arsenic-doped amorphous or polysilicon), (iii) oxide layer 204c, (iv) N.sup.+-doped amorphous silicon or polysilicon layer 204d (e.g., phosphorus or arsenic-doped amorphous or polysilicon), (v) conductor layer 204e (e.g., a titanium nitride-lined tungsten layer), channel layer 250 (e.g., any channel region formed out of any suitable semiconductor material described above), charge storage layer 251 (e.g., a multi-layer that may include any tunneling layer, any charge-trapping layer and any blocking layer, described above), and gate electrode or local word line 252 (e.g., any gate electrode described above). N.sup.+-doped amorphous silicon or polysilicon layers 204b and 204d extend lengthwise along the Y-direction to form, respectively, the common source region and the common drain region (common bit line) for all the storage transistors of the NOR memory string. Conductor layers 204a and 204e are in contact with and are provided to reduce the resistivities of the common source region and the common bit line, respectively.

(59) 3-dimensional array 1300 of NOR memory strings may be formed using any of the processes, or any combination thereof, discussed in Provisional Application III or Provisional Application IV (e.g., the process discussed in conjunction with FIGS. 2a-2j of Provisional Application III).

(60) The present inventors realize that the materials for the charge-trapping layer of the above-disclosed thin-film storage transistors (e.g., charge trapping layer 503 of FIG. 5), such as hafnium oxide, may have ferroelectric polarization phases, as known in the prior art. The present inventors realize that, by harnessing these ferroelectric phases for data storage, the thin-film storage transistors in a 3-dimensional memory array of NOR memory strings may be readily adapted to operate as ferroelectric field-effect transistors (FeFETs), thereby providing high endurance, long data retention, and relatively low voltage operations for both erase (under 7.0 volts) and program (under 7.0 volts). The combining of the ferroelectric polarization characteristics of the FeTFT and the 3-dimensional organization of the thin film horizontal (or, alternatively, vertical) NOR memory strings described herein achieves the additional benefits of high-density, low-cost memory arrays with thin film FeTFTs that can be randomly accessible at high speed (i.e. low read latency).

(61) FIG. 13(b) shows active stacks 1351-1 and 1351-2 of NOR memory strings in a 3-dimensional array 1350, with each NOR memory string including numerous FeFETs as storage transistors, according to one embodiment of the present invention. In 3-dimensional array 1350, each active stack (e.g., stack 1351-1) includes numerous NOR memory strings formed by the FeFETs (e.g., represented by NOR memory strings 254-1 to 254-4). The inset of FIG. 13(b) shows a cross-section of FeFET 1353 in one NOR memory string of active stack 1351-2.

(62) FIG. 13(b) shows representative FeFET 1353 in a NOR memory string of 3-dimensional array 1350. Representative FeFET 1353 includes (i) conductor layer 204a (e.g., a titanium nitride-lined tungsten layer), (ii) N.sup.+-doped amorphous silicon or polysilicon layer 204b, (iii) oxide layer 204c, (iv) N.sup.+-doped amorphous silicon or polysilicon layer 204d, and (v) conductor layer 204e (e.g., a titanium nitride-lined tungsten layer), which function in like manner, and may be provided in substantially the same manner, as those like layers that are assigned like reference numerals in FIG. 13(a). Rather than charge storage layer 251 of storage transistor 1303, FeFET 1353, however, has ferroelectric storage layer 271 that may include a ferroelectric material and an interface dielectric layer. FeFET 1353 has channel region 270 and gate electrode or local word line 272, which may be formed out of the same or different materials from channel region 250 and gate electrode or local word line 252, respectively. As in the storage transistors of a NOR memory string of FIG. 13(a), N.sup.+-doped amorphous silicon or polysilicon layers 204b and 204d extend lengthwise along the Y-direction to form, respectively, the common source region and the common drain region (common bit line) for all the FeFETs of the NOR memory string of FIG. 13(b). Likewise, conductor layers 204a and 204e in FIG. 13(b) are in contact with and are provided to reduce the resistivities of the common source region and the common bit line, respectively.

(63) In this detailed description, in all the embodiments of the present invention, the semiconductor substrate typically includes control, sensing and driving circuit that support the memory operations of the storage transistors or the FeFETs in the 3-dimensional array of NOR memory strings above it.

(64) In some embodiments, in order to reduce interference between neighboring FeFETs, ferroelectric storage layer 271 of FeFET 1353 of FIG. 13(b) is preferably separated from the ferroelectric storage layers of the FeFETs in other active multi-layers of its active stack, unlike charge storage layer 251 of storage transistor 1303 of FIG. 13(a), which may be continuous with the charge storage layers of the storage transistors in other active multi-layers of its active stack.

(65) According to one embodiment of the present invention, channel region 270 of FeFET 1350, which may be formed in a 3-dimensional memory array, may include p.sup.-doped poly silicon (e.g., 7.0-14.0 nm thick) and gate electrode 272 may be formed out of tungsten (W), molybdenum (Mo), aluminum (Al), ruthenium (Ru), tantalum (Ta), titanium (Ti), or any combination or alloy of these metals. Ferroelectric storage layer 271 may include an interface dielectric layer (e.g., silicon oxynitride (SiON), silicon nitride (Si.sub.3N.sub.4), or silicon oxide (SiO.sub.2), 0.0 to 2.0 nm thick, with a refractive index between 1.5 to 2.0) and a ferroelectric material layer (e.g., zirconium-doped hafnium oxide (HfO.sub.2:Zr; or HZO), aluminum-doped hafnium oxide (HfO.sub.2:Al), silicon-doped hafnium oxide (HfO.sub.2:Si) or lanthanum-doped hafnium oxide (HfO.sub.2:La). The ferroelectric material layer may be, for example, 3.0 to 8.0 nm thick. The term HZO may encompass hafnium zirconium oxide (HfZrO), hafnium zirconium oxynitride (HfZrON), hafnium zirconium aluminum oxide (HfZrAlO) or any hafnium oxide that includes zirconium impurities. The HZO ferroelectric material layer may be formed using atomic layer deposition (ALD) techniques at temperature between 200 C. to 330 C. (e.g., around 300 C.), with a post-deposition annealing step at a temperature between 400 C. and 1000 C., based on the desired crystallization phase requirement of the ferroelectric material.

(66) The interface dielectric layer isolates the ferroelectric material layer from electron or hole tunneling from the channel region during conduction, as the electrons or holes tunneled into the ferroelectric material layer may adversely affect polarization in the ferroelectric material layer. The interface dielectric layer may be formed out of a material with a dielectric constant greater than that of silicon oxide (high-k material; preferably with a dielectric constant greater than 3.9) to reduce the electric field during program or erase operations and to reduce the tunneling from the channel region. For a 0.0 nm thick interface dielectric layer, the ferroelectric material layer is directly deposited by atomic layer deposition (ALD) onto the channel region (e.g., polysilicon). A native oxide of a self-limiting thickness (e.g., 1.0 to 10.0 angstrom) would be inherently formed at the interface between the channel region and the ferroelectric material layer. This approach is particularly advantageous when the channel region is formed after high temperature steps, such that contamination by dopant diffusion is a lesser concern. In some embodiments, the bandgap-engineered tunneling layer (e.g., a silicon oxide (SiO.sub.2) and zirconium oxide (ZrO.sub.2) multi-layer) may serve as the interface dielectric layer, providing the advantage of reduced tunneling into the ferroelectric material layer. The high-k dielectric properties of the zirconium oxide reduce the electric field in the interface dielectric layer.

(67) An FeFET may be polarized to either a conducting or erased state or a non-conducting state or programmed state. In an FeFET, its threshold voltage in the erased state is lower than its threshold voltage in the conducting state. FIG. 14(a) shows hysteresis in the drain current (Ia) in response to an applied gate voltage (V.sub.g) in a conventional FeFET. (The conventional FeFET is formed at the planar surface of a monocrystalline semiconductor substrate and is not formed as a thin-film field effect transistor.) In FIG. 14(a), waveform 1401 traces the drain current of the FeFET in its erased state, as the gate voltage increases from less than 1.0 volts to greater than 1.0 volts, and waveform 1402 traces the drain current of the FeFET in its programmed state, as the gate voltage decreases from greater than 1.0 volts to less than 1.0 volts. As seen in FIG. 14(a), the conventional FeFET has a negative threshold voltage (V.sub.t).

(68) It is desirable in some applications, however, for an FeFET (e.g., a thin-film FeFET in a NOR memory string) to have a positive threshold voltage (V.sub.t), such as around 0.5 volts, to prevent undesirable leakage current when it is subject to a disturb condition (e.g., a neighboring FeFET in the NOR memory string, not the FeFET itself, is selected in a read operation).

(69) FIG. 14(b) shows a desirable hysteresis in the drain current (Ia) response to an applied gate voltage (V.sub.g) in a thin-film FeFET in a NOR memory array, according to one embodiment of the present invention. In FIG. 14(b), waveform 1403 traces the drain current of the FeFET in its erased state, as the gate voltage increases from less than 1.0 volts to greater than 1.0 volts, and waveform 1404 traces the drain current of the FeFET in its programmed state, as the gate voltage decreases from greater than 1.0 volts to less than 1.0 volts. As seen in FIG. 14(b), the FeFET has a positive threshold voltage (V.sub.t) of about 0.5 volts, and a threshold voltage difference (window) between the erased state and the programmed state of 1.0 volts to 2.5 volts. With a p.sup. polysilicon channel region (e.g., boron-doped), this threshold voltage may be achieved by (i) increasing the boron dopant concentration in the channel region, (ii) providing a gate electrode that is formed out of a conductive material that has a high work function (e.g., tungsten (W), molybdenum (Mo), aluminum (Al), ruthenium (Ru), tantalum (Ta), or titanium (Ti), (iii) appropriate biasing in the common source region (see below), or (iv) the combination of (i), (ii) and (iii).

(70) Table 1 summarizes exemplary bias voltages for a 3-dimensional array of NOR memory strings during erase, program and read operations at (i) the gate electrode or word line, the common source line, and the common bit line of a selected FeFET and (ii) the non-selected word lines and bit lines of the 3-dimensional memory array:

(71) TABLE-US-00002 TABLE 1 Selected Non-selected Gate/Word Word Source or Operation Line Source Drain Lines Bit Lines Erase 2.0 to 6.0 0 volts 0 volts e.g., 1.0 1.0 to 3.0 volts (e.g., volt volts (e.g., 4.0 volts) 2.0 volts) Program 0 volts 3.0 to 6.0 3.0 to 6.0 1.0 to e.g., 2.0 volts volts 3.0 volts volts (e.g., 4.0 (e.g., 4.0 (e.g., 1.0 volts) volts) volts) Read 1.0 to 3.0 0.0 volts 0.05 to 0.0 volts 0.0 to 1.0 volts 1.0 volts volts

(72) When the body region of an FeFET in a NOR memory string is floating, its programming speed may be slower than its erase speed. In such a condition, the gate-induced drain leakage (GIDL) effect may be harnessed to improve the programming speed. The GIDL effect may be activated by creating a voltage difference of 0.5 to 2.0 volts between the common bit line and the common source line (V.sub.ds) during programming, for example, by first momentarily pre-charging the common source line to a predetermined source line voltage through the common bit line, and then setting the common bit line to its target voltage, as disclosed in the Related Application.

(73) During a read operation, when the thin film FeFET in the erased state has a negative threshold voltage, its common source line may be biased to a voltage above such a threshold voltage to prevent conduction in the non-selected FeFETs of the NOR memory string. Of importance, during a read operation, the voltage between the gate or word line and the common bit line or the common source line in the selected FeFET is maintained at less than the voltage that may change the polarization phase of the selected FeTET, so as to avoid the phenomenon known as read-disturb.

(74) 3-dimensional memory array of thin film FeFET transistors that are organized as NOR memory strings may be formed by adapting any suitable process or processes disclosed in Provisional Applications III and IV.

(75) FIGS. 15(a)-15(d) illustrate a first process that forms a 3-dimensional memory array of thin film FeFETs organized as NOR memory strings, according to one embodiment of the present invention. FIG. 15(a) shows in top (i.e., X-Y plane) and X-Z plane cross-sectional view intermediate memory structure 1500, after formation of active stacks 1501-1 and 1501-2 and deposition into recesses of multi-layers 251-1 to 251-4 and etching back of a channel material (e.g., p.sup.-doped amorphous silicon or polysilicon). In FIG. 15(a), adjacent active stacks are separated by trenches 1502 that extend along the Y-direction. Intermediate memory structure 1500 may be achieved using the process steps, for example, illustrated by FIGS. 2a to 2h in Provisional Application III. In FIG. 15(a), channel region 270 in multi-layer 251-4 is separated to the like channel region of multi-layer 251-3 due to the multi-layers being recessed into adjacent isolation dielectric layers 203 (e.g., SiOC). Thereafter, trenches 1502 are filled by dielectric material 1504 (e.g., silicon oxide) and excess dielectric material is removed from the top of intermediate memory structure 1500 by, for example, chemical-mechanical polishing (CMP). Resulting intermediate structure 1500 is shown in FIG. 15(b).

(76) Thereafter, shafts 1505 (e.g., oval shafts) are formed in dielectric material 1505 of filled trenches 1502, using a process such as described in conjunction with FIG. 2j in Provisional Application IV. Resulting intermediate structure 1500 is shown in FIG. 15(c). Shafts 1505 each exposes sidewalls of isolation dielectric layers 203 and channel material 270.

(77) Thereafter, self-assembled monolayers (SAMs; e.g., species having active hydroxyl (OH) bonds) are provided to passivate the sidewalls of isolation dielectric layers 203. Thereafter, ferroelectric storage layer 271 may then be selectively deposited on the exposed surfaces of channel material 270. (The treatment by SAMs prevents deposition of ferroelectric storage layer 271 onto the sidewalls of isolation dielectric layers 203.) Resulting intermediate structure 1500 is shown in FIG. 15(d). Ferroelectric storage layer 271 may be formed by the selective deposition, using ALD techniques in an ozone ambient, of the interface dielectric layer and the ferroelectric material layer.

(78) The interface dielectric layer may include a native oxide layer formed by a chemical clean of the surface of channel material 270, followed by densification, for example by pulsed ozone or by thermal annealing in a hydrogen or deuterium ambient, or any other techniques known to a person of ordinary skill in the art. This treatment reduces electronic leakage through the interface dielectric layer, reduces the surface states at the interface between the third semiconductor layer and the ferroelectric storage layer, or both. The ferroelectric material layer may be formed for example, using repeated cycles of hafnium oxide depositions and zirconium oxide depositions (e.g., in a HfO.sub.2:ZrO.sub.2 ratio of 4:1). For the thicker ferroelectric material layers (e.g., greater than 40 nm), additional SAM treatments between deposition cycles may be advisable. FIG. 15(d) includes an additional X-Y plane cross-sectional view 1510 of memory structure 1500 along line A-A through oxide layer 204c of the X-Y plane cross sectional view. The 3-dimensional array may then be completed using, for example, the process steps described in conjunction with FIGS. 2l to 2t of Provisional Application IV.

(79) FIGS. 16(a)-16(d) illustrate a second process that forms a 3-dimensional memory array of thin film FeFETs organized as NOR memory strings, according to one embodiment of the present invention. FIG. 16(a) shows an X-Z plane cross-sectional view intermediate memory structure 1600, after formation of active stacks 1601-1 and 1601-2 and deposition into recesses of multi-layers 251-1 to 251-4 and etching back of a channel material (e.g., p.sup.-doped amorphous silicon or polysilicon) to expose the sidewalls of isolation dielectric layer 203. As in FIG. 15(a), adjacent active stacks in FIG. 16(a) are separated by trenches (i.e., trenches 1602) that extend along the Y-direction. Intermediate memory structure 1600 may be achieved using the process steps, for example, illustrated by FIGS. 2a to 2h in Provisional Application III.

(80) Relative to intermediate memory structure 1500 of FIG. 15(a), however, the recesses of multi-layers 251-1 to 251-4 in FIG. 16(a) are deeper. For example, for channel region 270 targeted for a 10.0 nm thickness, the recesses of multi-layers 251-1 to 251-4 are made 20.0 nm thick, such that channel material 270 is 20.0 nm thick at the process step of FIG. 16(a). A further etch back of channel region 270 (e.g., by a wet etch) reduces the thickness of channel regions 270, for example, to 10.0 nm, thereby creating recesses of approximately 10.0 nm deep between adjacent isolation dielectric layers 203. Resulting intermediate memory structure 1600 is shown in FIG. 16(b).

(81) Thereafter, ferroelectric storage layer 271 may be formed by the deposition, using ALD techniques, of the interface dielectric layer and the ferroelectric material layer onto channel material 270 in the 10.0 nm deep recesses of between adjacent isolation dielectric layers 203. The ferroelectric material layer may be formed for example, using repeated cycles of hafnium oxide depositions and zirconium oxide depositions (e.g., in a HfO.sub.2:ZrO.sub.2 ratio of 4:1). Resulting intermediate memory structure 1600 is shown in FIG. 16(c). As shown in FIG. 16(c), because of the recesses, the ferroelectric storage layer 271 of the multi-layers are separated from each other by isolation dielectric layers 203.

(82) The 3-dimensional array of NOR memory strings may then be completed using, for example, the process steps described in conjunction with FIGS. 2j to 2t of Provisional Application IV. FIG. 16(d) shows a X-Z plane cross-sectional view of the completed 3-dimensional array of NOR memory strings, showing (i) conductive material 272 (i.e., gate electrode) in shaft 1605 and (ii) oxide 1604 electrically isolating adjacent gate electrodes from each other.

(83) FIGS. 17(a)-17(g) illustrate a third process that forms a 3-dimensional memory array of thin film FeFETs organized as NOR memory strings, according to one embodiment of the present invention. FIG. 17(a) shows an X-Z plane cross-sectional view intermediate memory structure 1700, after formation of active stacks 1701-1 and 1701-2 and deposition into recesses of multi-layers 251-1 to 251-4 and etching back of a channel material (e.g., p.sup.-doped amorphous silicon or polysilicon; e.g., 20.0 nm thick) to expose the sidewalls of isolation dielectric layer 203. As in FIG. 16(a), adjacent active stacks in FIG. 17(a) are separated by trenches (i.e., trenches 1702) that extend along the Y-direction. A further etch back of channel region 270 (e.g., by a wet etch) reduces the thickness of channel regions 270, for example, to 10.0 nm, thereby creating recesses of approximately 10.0 nm deep between adjacent isolation dielectric layers 203. Resulting intermediate memory structure 1700 is shown in FIG. 17(b). Intermediate memory structures 1700 of FIGS. 17(a) and 17(b) may be formed using substantially the same process steps as those discussed above that are used to form intermediate memory structure 1600 of FIGS. 16(a) and 16(b), respectively.

(84) Thereafter, ferroelectric storage layer 271 is formed on intermediate memory structure 1700, using the ALD techniques discussed, for example, in conjunction with FIG. 16(c) above, followed by a conformal deposition of amorphous silicon liner 1707. Resulting intermediate memory structure 1700 is shown in FIG. 17(c). Then, an anisotropic dry etch step removes portions of amorphous silicon liner 1707, exposing the portions of ferroelectric storage layer 271 on the sidewalls of isolation dielectric layers 203 in trenches 1702, while allowing the remainder of amorphous silicon liner 1707 protect the portions of ferroelectric storage layer 271 in the recesses of multi-layers 251-1 to 251-4. The anisotropic dry etch also sputters away amorphous silicon liner 1707 and ferroelectric storage layer 271 from top of intermediate memory structure 1700. Resulting intermediate memory structure 1700 is shown in FIG. 17(d).

(85) Thereafter, a wet etch that removes the ferroelectric material layer (e.g., hafnium zirconium oxide (HZO)) removes ferroelectric storage layer 271 from the sidewalls of isolation dielectric layers 203. Resulting intermediate memory structure 1700 is shown in FIG. 17(e). Then, the remainder of amorphous silicon liner 1707 may be removed by a wet etch. Resulting intermediate memory structure 1700 is shown in FIG. 17(f).

(86) The 3-dimensional array of NOR memory strings may then be completed using, for example, the process steps described in conjunction with FIGS. 2j to 2t of Provisional Application IV. FIG. 17(g) shows a X-Z plane cross-sectional view of the completed 3-dimensional array of NOR memory strings, showing (i) conductive material 272 (i.e., gate electrode) in shaft 1705, and (ii) oxide 1704 electrically isolating adjacent gate electrodes from each other.

(87) According to another embodiment of the present invention, channel region 270 of FeFET 1353 may be formed with an 8.0 to 15.0 nm thick oxide semiconductor material (e.g., indium zinc oxide (InZnO; or IZO). An IZO channel region has the advantage of a high mobility for greater switching performance and without concern for electron or hole tunneling. For example, a 10.0 nm thick IZO film has an electron mobility of 40.6 cm.sup.2/V, relative to an aluminum zirconium oxide (AZO) of comparable thickness with an electron mobility of 5.6 cm.sup.2/V. Furthermore, the common source region and the common bit line may be formed out of a metal (e.g., Mo). The ferroelectric storage layer of FeFET 1353 may be provided by any of the ferroelectric storage described above (e.g., an SiON interface dielectric layer and a HZO ferroelectric material layer.) As this FeFET does not have a p/n junction, any leakage current from an FeFET in the programmed state is relatively small. Consequently, such an FeFET is particularly advantageous for high-temperature applications. Such an FeFET may also be built with a relatively shorter channel length, as there is no need for a margin to allow for dopant diffusion from the heavily-doped semiconductor common bit line and common source line during any annealing step affecting the channel region. The metal common bit line and common source line also reduces the thickness of the active multi-layer (e.g., 40.0 nm common bit line or common source line, 40.0 nm channel region, and 30.0 nm SiOC interlayer dielectric, for a total of a relatively thinner 150.0 nm). The common source and drain region may be built using a sacrificial material that is replaced in a late metal-replacement step.

(88) The 3-dimensional horizontal NOR memory strings of FeFETs disclosed herein have a significant advantages in that they offer a relatively large surface area for the ferroelectric storage layer (e.g., ferroelectric storage layer 271 in FIG. 15(d), 16(d) or 17(j)) when built in the 3-dimensional memory structures disclosed herein (e.g., memory structures 1300, 1500, 1600 or 1700), while requiring a very small footprint on the semiconductor substrate by virtue of their vertical orientation relative to the substrate. This larger surface area provides a tight distribution of voltages in the erase and program states, which are difficult to achieve in highly-scaled planar FeTET.

(89) The FeFETs disclosed herein are illustrated above by adapting storage transistors in the 3-dimensional horizontal NOR memory strings, such as those disclosed in Provisional Applications III-IV. However, FeFETs may also be formed by adapting storage transistors in 3-dimensional vertical NOR memory strings, such as those disclosed in Provisional Application V, by applying substantially the same principles and methods disclosed herein.

(90) The detailed description above is provided to illustrate specific embodiments of the present invention and is not intended to be limiting. Numerous variations and modifications within the scope of the present invention are possible. For example, while the detailed description above illustrates the present invention using thin-film field-effect transistors with PN junctions between semiconductor layers (e.g., polysilicon layers), the present invention can also be applicable to junction-less transistors. In some embodiments, such junction-less transistors may include thin-film junction-less transistors with a conductive oxide channel region. In some embodiments, suitable conductive metal oxides include gallium oxides, zinc oxides, indium oxides (e.g., indium gallium zinc oxide (IGZO) and indium zinc oxide (IZO)) and any suitable conductive metal oxides with charge-carriers having mobilities can be modified or modulated using suitable preparation or inclusion of suitable impurities. For example, in one embodiment, instead of the polysilicon thin-film field effect transistor with its N.sup.+ polysilicon source and drain regions and P.sup. polysilicon channel region, a junction-less transistor with a low-resistivity conductive material (e.g., titanium nitride (TiN)-lined tungsten (W), tungsten, cobalt, molybdenum) providing the source and drain regions, and a conductive metal oxide (e.g., IGZO) providing a channel region.

(91) The present invention is set forth in the accompanying claims.