SEMICONDUCTOR DEVICE

Abstract

A product-sum calculation with high power efficiency is performed while maintaining a small area of a memory cell. A semiconductor device includes a memory cell array in which a plurality of memory cells is arranged in a matrix. Then, each memory cell of the plurality of memory cells includes a flip-flop circuit including two inverter circuits in each of which a load field effect transistor and a drive field effect transistor are connected in series, input portions and output portions of the two inverter circuits being cross-joined to each other, two transfer field effect transistors each having a gate electrode connected to a word line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the output portions of the two inverter circuits, and two resistance elements of which one end sides are respectively connected to the second main electrode regions of the two transfer field effect transistors and other end sides are respectively connected to a bit line and a bit line bar.

Claims

1. A semiconductor device comprising: a memory cell array in which a plurality of memory cells is arranged in a matrix, wherein each memory cell of the plurality of memory cells includes a flip-flop circuit including two inverter circuits in each of which a load field effect transistor and a drive field effect transistor are connected in series, input portions and output portions of the two inverter circuits being cross-joined to each other, two transfer field effect transistors each having a gate electrode connected to a word line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the output portions of the two inverter circuits, and two resistance elements of which one end sides are respectively connected to the second main electrode regions of the two transfer field effect transistors and other end sides are respectively connected to a bit line and a bit line bar.

2. The semiconductor device according to claim 1, wherein a resistance value of the resistance elements is greater than a channel resistance value of the transfer field effect transistors.

3. The semiconductor device according to claim 1, wherein a resistance value of the resistance elements is 1 MΩ or more.

4. The semiconductor device according to claim 1, wherein when a product-sum calculation is performed, the word line functions as an axon, the bit line functions as a dendrite, and the bit line bar functions as a dendrite bar.

5. A semiconductor device comprising: a memory cell array in which a plurality of memory cells is arranged in a matrix, wherein each memory cell of the plurality of memory cells includes a flip-flop circuit including two inverter circuits in each of which a load field effect transistor and a first drive field effect transistor are connected in series, input portions and output portions of the two inverter circuits being cross-joined to each other, two first transfer field effect transistors each having a gate electrode connected to a word line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the output portions of the two inverter circuits, and the second main electrode regions being respectively connected to a bit line and a bit line bar, two second drive field effect transistors each having a gate electrode and a pair of first and second main electrode regions, the gate electrodes being respectively connected to the input portions of the two inverter circuits, and the first main electrode regions being connected to each other, two second transfer field effect transistors each having a gate electrode connected to the word line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the second main electrode regions of the two second drive field effect transistors, and two resistance elements of which one end sides are respectively connected to the second main electrode regions of the two second transfer field effect transistors and other end sides are respectively connected to a dendrite line and a dendrite line bar.

6. The semiconductor device according to claim 5, wherein a resistance value of the resistance elements is greater than a channel resistance value of the transfer field effect transistors.

7. The semiconductor device according to claim 5, wherein a resistance value of the resistance elements is 1 MΩ or more.

8. The semiconductor device according to claim 5, wherein when a product-sum calculation is performed, the word line functions as an axon.

9. A semiconductor device comprising: a memory cell array in which a plurality of memory cells is arranged in a matrix, wherein each memory cell of the plurality of memory cells includes a flip-flop circuit including two inverter circuits in each of which a load field effect transistor and a first drive field effect transistor are connected in series, input portions and output portions of the two inverter circuits being cross-joined to each other, a first transfer field effect transistor having a gate electrode connected to a word line, and a pair of first and second main electrode regions, the first main electrode region being connected to the output portion of the other inverter circuit of the two inverter circuits, and the second main electrode region being connected to a bit line bar, two second drive field effect transistors each having a gate electrode and a pair of first and second main electrode regions, the gate electrodes being respectively connected to the input portions of the two inverter circuits, and the first main electrode regions being connected to each other, two second transfer field effect transistors each having a gate electrode connected to an axon line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the second main electrode regions of the two second drive field effect transistors, and two resistance elements of which one end sides are respectively connected to the second main electrode regions of the two second transfer field effect transistors and other end sides are respectively connected to a dendrite line and a dendrite line bar.

10. The semiconductor device according to claim 9, wherein a resistance value of the resistance elements is greater than a channel resistance value of the transfer field effect transistors.

11. The semiconductor device according to claim 9, wherein a resistance value of the resistance elements is 1 MΩ or more.

12. A semiconductor device comprising: a memory cell array in which a plurality of memory cells is arranged in a matrix, wherein each memory cell of the plurality of memory cells includes a flip-flop circuit including two inverter circuits in which a load field effect transistor and a first drive field effect transistor are connected in series, input portions and output portions of the two inverter circuits being cross-joined to each other, two first transfer field effect transistors each having a gate electrode connected to a word line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the output portions of the two inverter circuits, and the second main electrode regions being respectively connected to a bit line and a bit line bar, two second drive tunnel field effect transistors each having a gate electrode and a pair of n-type first main electrode region and p-type second main electrode region, the gate electrodes being respectively connected to the input portions of the two inverter circuits, and the n-type first main electrode region being connected to an axon line, and two second transfer tunnel field effect transistors each having a gate electrode connected to the word line, and a pair of n-type first main electrode region and p-type second main electrode region, the p-type second main electrode regions being respectively connected to the p-type second main electrode regions of the two second drive tunnel field effect transistors, and the n-type first main electrode regions being respectively connected to a dendrite line and a dendrite line bar.

Description

BRIEF DESCRIPTION OF DRAWINGS

[0020] FIG. 1 is a diagram illustrating a schematic configuration of a memory cell array unit of a semiconductor device according to a first embodiment of the present technology.

[0021] FIG. 2 is an equivalent circuit diagram of the memory cell in FIG. 1.

[0022] FIG. 3 is a schematic plan view illustrating a planar pattern of the memory cell in FIG. 1.

[0023] FIG. 4A is a schematic sectional view illustrating a sectional structure taken along the line a3-a3 in FIG. 3.

[0024] FIG. 4B is a schematic sectional view illustrating a sectional structure taken along the line b3-b3 in FIG. 3.

[0025] FIG. 5 is a diagram illustrating a schematic configuration of a memory cell array unit of a semiconductor device according to a second embodiment of the present technology.

[0026] FIG. 6 is an equivalent circuit diagram of the memory cell in FIG. 5.

[0027] FIG. 7 is a diagram illustrating a schematic configuration of a memory cell array unit of a semiconductor device according to a third embodiment of the present technology.

[0028] FIG. 8 is an equivalent circuit diagram of the memory cell in FIG. 7.

[0029] FIG. 9 is a diagram illustrating a schematic configuration of a memory cell array unit of a semiconductor device according to a fourth embodiment of the present technology.

[0030] FIG. 10 is an equivalent circuit diagram of the memory cell in FIG. 9.

[0031] FIG. 11 is a transmissive circuit diagram illustrating a modification example of the memory cell according to the fourth embodiment.

[0032] FIG. 12 is a transmissive circuit diagram illustrating a modification example of the memory cell according to the fourth embodiment.

MODE FOR CARRYING OUT THE INVENTION

[0033] Hereinafter, embodiments of the present technology will be described in detail with reference to the drawings.

[0034] Note that, in all the drawings for describing the embodiments of the present technology, constituents having the same functions are denoted by the same reference numerals, and repeated description thereof will be omitted.

[0035] Furthermore, each drawing is schematic and may be different from an actual one. Furthermore, the following embodiments exemplify a device or a method for embodying the technical idea of the present technology, and do not specify the configuration as follows. That is, various modifications can be applied to the technical idea of the present technology within the technical scope described in the claims.

[0036] Furthermore, in the following embodiments, in three directions orthogonal to each other in a space, a first direction and a second direction orthogonal to each other in the same plane are defined as an X direction and a Y direction, respectively, and a third direction orthogonal to the first direction and the second direction is defined as a Z direction. In the following embodiments, a thickness direction of a semiconductor layer (semiconductor substrate) will be described as the Z direction.

First Embodiment

[0037] In a first embodiment, an SRAM type memory cell including six field effect transistors and two resistance elements will be described.

[0038] <<Configuration of Memory Cell Array Unit>>

[0039] A semiconductor device 1 according to the first embodiment of the present technology includes a memory cell array unit 2 illustrated in FIG. 1. As illustrated in FIG. 1, in the memory cell array unit 2, a plurality of memory cells 3 are arranged in a matrix on a two-dimensional plane including the X direction and the Y direction. Furthermore, in the memory cell array unit 2, word lines WL extending in the X direction are arranged for each of the memory cells 3 arranged in the Y direction. Furthermore, in the memory cell array unit 2, complementary bit lines (a bit line BL.sub.1 and a bit line bar BL.sub.2 (BL-)) extending in the Y direction are arranged for each of the memory cells 3 arranged in the X direction. Then, each memory cell 3 of the plurality of memory cells 3 is disposed at an intersection between the corresponding word line WL and the complementary bits (BL.sub.1 and BL.sub.2).

[0040] Here, during a product-sum calculation process (inference in a neural network), the memory cell array unit 2 functions as a product-sum calculation circuit, the word line WL functions as an axon, and the bit line BL.sub.1 and the bit line bar BL.sub.2 function as a dendrite and a dendrite bar (dendrite-).

[0041] <<Configuration of memory cell>>

[0042] As illustrated in FIG. 2, each of the plurality of memory cells 3 includes a flip-flop circuit 5, two transfer field effect transistors (transfer gate transistors) Qt.sub.1 and Qt.sub.2, and two resistance elements 6A and 6B.

[0043] The flip-flop circuit 5 includes two inverter circuits 4a and 4b, and has a configuration in which input portions 4a.sub.1 and 4b.sub.1 and the output portions (storage node portions) 4a.sub.2 and 4b.sub.2 of two inverter circuits 4a and 4b are alternately cross-joined.

[0044] One inverter circuit 4a of the two inverter circuits 4a and 4b has a configuration in which a load field effect transistor (pull-up transistor) Qp.sub.1 and a drive field effect transistor (pull-down transistor) Qd.sub.1 are connected in series. The other inverter circuit 4b has a configuration in which a load field effect transistor Qp.sub.2 and a drive field effect transistor Qd.sub.2 are connected in series.

[0045] The two load field effect transistors Qp.sub.1 and Qp.sub.2, the two drive field effect transistors Qd.sub.1 and Qd.sub.2, and the two transfer field effect transistors Qt.sub.1 and Qt.sub.2 each have a gate insulating film, a gate electrode (control electrode), and a pair of first main electrode region and second main electrode region functioning as a source region and a drain region, and electrical conduction between the first main electrode region and the second main electrode region is controlled by a gate signal input to the gate electrode. Although the field effect transistors Qp.sub.1, Qp.sub.2, Qd.sub.1, Qd.sub.2, Qt.sub.1, and Qt.sub.2 are not limited thereto, for example, the field effect transistors Qp.sub.1, Qp.sub.2, Qd.sub.1, and Qd.sub.2 are configured by MOSFETs in which a gate insulating film is a silicon oxide (SiO.sub.2) film. Furthermore, as the field effect transistors Qp.sub.1, Qp.sub.2, Qd.sub.1, Qd.sub.2, Qt.sub.1, and Qt.sub.2, a metal insulator semiconductor FET (MISFET) in which a gate insulating film is a silicon nitride (Si.sub.3N.sub.4) film or a laminated film such as a silicon nitride film and a silicon oxide film may be used. Hereinafter, the load field effect transistor may be simply referred to as a load FET, the drive field effect transistor may be simply referred to as a drive FET, and the transfer field effect transistor may be simply referred to as a transfer FET.

[0046] The two load FETs Qp.sub.1 and Qp.sub.2 are configured by p-channel conductivity type MOSFETs. On the other hand, the two drive FETs Qd.sub.1 and Qd.sub.2 and the two transfer FETs Qt.sub.1 and Qt.sub.2 are configured by n-channel conductivity type MOSFTETs. That is, the memory cell 3 includes a CMOS circuit.

[0047] As illustrated in FIG. 2, in the one inverter circuit 4a, the gate electrodes of the load FET Qp.sub.1 and the drive FET Qd.sub.1 are electrically connected to each other to configure an input portion 4a.sub.1. Furthermore, the respective first main electrode regions (drain regions) of the load FET Qp.sub.1 and the drive FET Qd.sub.1 are electrically connected to each other to configure an output portion 4a.sub.2. Furthermore, the second main electrode region (source region) of the drive FET Qd.sub.1 is electrically connected to a ground line 28c.sub.1 (refer to FIG. 3) to which a Vss potential (for example, 0 V) as a first reference potential is applied. Furthermore, the second main electrode region (source region) of the load FET Qp.sub.1 is electrically connected to a power supply line 28d (refer to FIG. 3) to which a Vdd potential (for example, 0.5 V to 1.2 V) as a second reference potential higher than the Vss potential as the first reference potential is applied.

[0048] As illustrated in FIG. 2, in the other inverter circuit 4b, the gate electrodes of the load FET Qp.sub.2 and the drive FET Qd.sub.2 are electrically connected to each other to configure an input portion 4b.sub.1. The first main electrode regions (drain regions) of the load FET Qp.sub.2 and the drive FET Qd.sub.2 are electrically connected to each other to configure an output portion 4b.sub.2. Then, the second main electrode region (source region) of the drive FET Qd.sub.2 is also electrically connected to a ground line 28c.sub.2 (refer to FIG. 3) to which the Vss potential is applied similarly to the drive FET Qd.sub.1, and the second main electrode region (source region) of the load FET Qp.sub.2 is also electrically connected to the power supply line 18d (refer to FIG. 3) to which the Vdd potential is applied similarly to the load FET Qp.sub.1.

[0049] As illustrated in FIG. 2, in the two inverter circuits 4a and 4b, the output portion 4a.sub.2 of one inverter circuit 4a is electrically connected to the input portion 4b.sub.1 of the other inverter circuit 4b. That is, the first main electrode region (drain region) of each of the load FET Qp.sub.1 and the drive FET Qd.sub.1 configuring one inverter circuit 4a is electrically connected to the gate electrode of each of the load FET Qp.sub.2 and the drive FET Qd.sub.2 configuring the other inverter circuit 4b.

[0050] Furthermore, the output portion 4b.sub.2 of the other inverter circuit 4b is electrically connected to the input portion 4a.sub.1 of the one inverter circuit 4a. That is, the first main electrode region (drain region) of each of the load FET Qp.sub.2 and the drive FET Qd.sub.2 configuring the other inverter circuit 4b is electrically connected to the gate electrode of each of the load FET Qp.sub.1 and the drive FET Qd.sub.1 configuring the one inverter circuit 4a.

[0051] Note that the respective output portions 4a.sub.2 and 4b.sub.2 of the two inverter circuits 4a and 4b configure a storage node portion of the memory cell 3.

[0052] In the two transfer FETs Qt.sub.1 and Qt.sub.2, in one transfer FET Qt.sub.1, the gate electrode is electrically connected to the word line WL, and the first main electrode region of the pair of first and second main electrode regions is electrically connected to the output portion 4a.sub.2 of one inverter circuit 4a. In the other transfer FET Qt.sub.2, the gate electrode is electrically connected to the word line WL, and the first main electrode region of the pair of first and second main electrode regions is electrically connected to the output portion 4b.sub.2 of the other inverter circuit 4b.

[0053] As illustrated in FIG. 2, in the two resistance elements 6A and 6B, one resistance element 6A has one end side electrically connected to the second main electrode region of one transfer FET Qt.sub.1 and the other end side electrically connected to the bit line BL.sub.1. In the other resistance element 6B, one end side is electrically connected to the second main electrode region of the other transfer FET Qt.sub.2, and the other end side is electrically connected to the bit line bar BL.sub.2.

[0054] <<Specific Configuration of Memory Cell>>

[0055] Next, a specific configuration of the memory cell 3 will be described in detail with reference to FIGS. 3, 4A, and 4B.

[0056] <Configuration of FFT>

[0057] In the one inverter circuit 4a, as illustrated in FIGS. 3 and 4A, the load FET Qp.sub.1 is configured in an n-type well region 12a corresponding to an n-type semiconductor region in an active region partitioned by an isolation region 11 on the main surface of the semiconductor layer 10. The isolation region 11 is not limited thereto, but is configured by, for example, a shallow trench isolation (STI) structure constructed by forming a groove extending in the depth direction from the main surface of the semiconductor layer 10 and selectively burying an insulating film in the groove.

[0058] The load FET Qp.sub.1 mainly includes a gate insulating film 15 provided on the main surface of the semiconductor layer 10, a gate electrode 16p provided on the gate insulating film 15, and a pair of first main electrode regions 17p.sub.1 and second main electrode regions 17p.sub.2provided in the semiconductor layer 10 to be separated from each other in the channel length direction with a channel formation region immediately below the gate electrode 16p interposed therebetween and functioning as a source region and a drain region. The pair of first and second main electrode regions 17p.sub.1 and 17p.sub.2 is provided in the n-type well region 12a.

[0059] In the inverter circuit 4a, as illustrated in FIGS. 3 and 4B, the drive FET Qd.sub.1 is configured in a p-type well region 12b from a p-type semiconductor region in an active region partitioned by the isolation region 11 on the main surface of the semiconductor layer 10. The drive FET Qd.sub.1 mainly includes a gate insulating film 15 provided on the main surface of the semiconductor layer 10, a gate electrode 16d provided on the gate insulating film 15, and a pair of first main electrode region 17d.sub.1 and second main electrode region 17d.sub.2 provided in the semiconductor layer 10 to be separated from each other in the channel length direction with a channel formation region immediately below the gate electrode 16d interposed therebetween and functioning as a source region and a drain region. The pair of first and second main electrode regions 17d.sub.1 and 17d.sub.2 is provided in the p-type well region 12b.

[0060] As illustrated in FIGS. 3 and 4B, one transfer FET Qt.sub.1 is configured in a p-type well region 12b in an active region defined by the isolation region 11 on the main surface of the semiconductor layer 10. One transfer FET Qt.sub.1 mainly includes a gate insulating film 15 provided on the main surface of the semiconductor layer 10, a gate electrode 16t provided on the gate insulating film 15, and a pair of first main electrode region 17t.sub.1 and second main electrode region 17t.sub.2 provided in the semiconductor layer 10 to be separated from each other in the channel length direction with a channel formation region immediately below the gate electrode 16t interposed therebetween and functioning as a source region and a drain region. The pair of first and second main electrode regions 17t.sub.1 and 17t.sub.2 is provided in the p-type well region 12b. The transfer FET Qt.sub.1 and the drive FET Qd.sub.1 have a structure in which the respective first main electrode regions 17t.sub.1 and 17d.sub.1 are made common (shared).

[0061] The semiconductor layer 10 is configured with, for example, a p-type semiconductor substrate including single crystal silicon. The gate insulating film 15 is configured with, for example, a silicon oxide (SiO.sub.2) film. The gate electrodes 16p, 16d, and 16t are configured with, for example, a composite film in which a silicide film is laminated on a polycrystalline silicon film into which an impurity for reducing a resistance value is introduced.

[0062] The pair of first and second main electrode regions 17p.sub.1 and 17p.sub.2 includes, for example, an extension region including a p-type semiconductor region, a contact region including a p-type semiconductor region having a higher impurity concentration than that of the extension region, and a silicide film provided on the contact region. The pair of first and second main electrode regions 17d.sub.1 and 17d .sub.2 and the pair of first and second main electrode regions 17t.sub.1 and 15t.sub.2 include, for example, an extension region including an n-type semiconductor region, a contact region including an n-type semiconductor region having a higher impurity concentration than that of the extension region, and a silicide film provided on the contact region.

[0063] As illustrated in FIGS. 4A and 4B, each of the load FET Qp.sub.1, the drive FET Qd.sub.1, and the transfer FET Qt.sub.1 is covered with an interlayer insulating film 21 provided on the semiconductor layer 10.

[0064] In the other inverter circuit 4b, the transfer FET Qt.sub.2 has a configuration similar to that of the above-described transfer FET Qt.sub.1, and the drive FET Qd.sub.2 has a configuration similar to that of the above-described drive FET Qd.sub.1. Then, the other transfer FET Qt.sub.2 has a configuration similar to that of the above-described transfer FET Qt.sub.1. Therefore, a description of specific configurations of the load FET Qt.sub.2, the drive FET Qd.sub.2, and the transfer FET Qt.sub.2 will be omitted.

[0065] Furthermore, each of the load FETs Qp.sub.1 and Qp.sub.2, the drive FETs Qd.sub.1 and Qd.sub.2, and the transfer FETs Qt.sub.1 and Qt.sub.2 is configured by, for example, a lightly doped drain (LDD) structure and a self-aligned silicide (SALICIDE) structure, but a description of a specific configuration thereof will also be omitted.

[0066] Note that the load FET Qp.sub.2, the drive FET Qd.sub.2, and the transfer FET Q.sub.2 are also covered with the interlayer insulating film 21.

[0067] <Configuration of Resistance Element>

[0068] As illustrated in FIG. 4B, one resistance element 6A is buried in an interlayer insulating film 24 provided on the interlayer insulating film 21. The resistance element 6A has, not limited to, a metal insulator metal (MIM) structure in which, for example, a first electrode 23a on one end side, an insulating film 23b, and a second electrode 23c on the other end side are laminated in this order from the semiconductor layer 10 side. As each of the first electrode 23a and the second electrode 23c, for example, a high melting point metal compound film such as a titanium nitride (TiN) film or a tantalum nitride (TaN) film may be used. As the insulating film 23b, for example, a silicon oxide (SiO.sub.2) film, an aluminum oxide (AlO.sub.2) film, a magnesium oxide (MgO.sub.2) film, a hafnium oxide (HfO.sub.2) film, or a zirconium oxide (ZrO.sub.2) film having a film thickness of about 1 to 3 nm can be used.

[0069] The other resistance element 6B has a configuration similar to that of the above-described resistance element 6A. Therefore, a description of a specific configuration of the other resistance element 6B will be omitted.

[0070] The two resistance elements 6A and 6B can obtain a desired resistance value with a small area. For example, the resistance elements 6A and 6B can form a resistance value of 1 MΩ or more, which is necessary in a case where a bit line as a summation line is charged with product-sum charge for multiple bits (for example, 1024), within an occupied area of the memory cell 3 during a product-sum calculation process. Moreover, since the resistance elements 6A and 6B are disposed on the transfer FETs Qt.sub.1 and Qt.sub.2 , in other words, disposed to overlap the transfer FETs Qt.sub.1 and Qt.sub.2 in a plan view (refer to FIGS. 3 and 4B), an increase in the planar size of the memory cell 3 can be suppressed. That is, the memory cell 3 of the first embodiment includes the two resistance elements 6A and 6B while maintaining a small area necessary for arrangement of six FETs (Q.sub.1, Qp.sub.2, Qd.sub.1, Qd.sub.2, Qt.sub.1, and Qt.sub.2).

[0071] A resistance value of each of the two resistance elements 6A and 6B is preferably larger than a channel resistance value of the transfer FETs Qt.sub.1 and Qt.sub.2, and more preferably 1 MΩ or more.

[0072] <Configuration of One Inverter Circuit>

[0073] As illustrated in FIG. 3, the gate electrodes 16p and 16d of the load FET Qp.sub.1 and the drive FET Qd.sub.1 configuring one inverter circuit 4a are integrally molded and electrically connected to each other. That is, the respective gate electrodes 16p and 16d of the load FET Qp.sub.1 and the drive FET Qd.sub.1 are connected to each other to configure the input portion 4a.sub.1 (refer to FIG. 2).

[0074] As illustrated in FIGS. 3 and 4A, the first main electrode region 17p.sub.1 of the load FET Qp.sub.1 is electrically connected to a relay wiring 25a.sub.1 formed in a first wiring layer on the interlayer insulating film 21 via a conductive plug 22a buried in the interlayer insulating film 21 on the semiconductor layer 10. On the other hand, as illustrated in FIGS. 3 and 4B, the respective first main electrode regions 17d.sub.1 and 17t.sub.1 of the drive FET Qd.sub.1 and the transfer FET Qt.sub.1 are electrically connected to the relay wiring 25a.sub.1 via a conductive plug 22b buried in the interlayer insulating film 21 on the semiconductor layer 10. That is, the respective first main electrode regions 17p.sub.1, 17d.sub.1 , and 17t.sub.1 of the load FET Qp.sub.1, the drive FET Qd.sub.1, and the transfer FET Qt.sub.1 are electrically connected to each other to configure the output portion 4a.sub.2 (refer to FIG. 2). The relay wiring 25a.sub.1 and relay wirings 25c.sub.1 to 25e.sub.1 that will be described later are buried in an interlayer insulating film 24 on the interlayer insulating film 21.

[0075] As illustrated in FIGS. 3 and 4B, the second main electrode region 17d.sub.2 of the drive FET Qd.sub.1 is electrically connected to the relay wiring 25c.sub.1 formed in the first wiring layer via a conductive plug 22c buried in the interlayer insulating film 21. Then, the relay wiring 25c.sub.1 is electrically connected to a ground wiring 28c.sub.1 formed in a second wiring layer on the interlayer insulating film 26 and extending in the Y direction via a conductive plug 27c buried in the interlayer insulating film 26 on the interlayer insulating film 24. The Vss potential as the above-described first reference potential is applied to the ground wiring 28c.sub.1. That is, Vss potential is supplied from the ground wire 28c.sub.1 to the second main electrode region 17d.sub.2 of the drive FET Qd.sub.1.

[0076] As illustrated in FIGS. 3 and 4A, the second main electrode region 17p.sub.2 of the transfer FET Qp.sub.1 is electrically connected to the relay wiring 25d.sub.1 formed in the first wiring layer via a conductive plug 22d buried in the interlayer insulating film 21. Then, although not illustrated in detail, the relay wiring 25d.sub.1 is electrically connected to the power supply line 28d formed in the second wiring layer on the interlayer insulating film 26 and extending in the Y direction via the conductive plug buried in the interlayer insulating film 26. The Vdd potential as the above-described second reference potential is applied to the power supply line 28d. That is, the Vdd potential is supplied from the power supply line 28d to the second main electrode region 17p.sub.2 of the load FET Qp.sub.1.

[0077] As illustrated in FIGS. 3 and 4B, the second main electrode region 17t.sub.2 of the transfer FET Qt.sub.1 is electrically connected to the first electrode 23a of the resistance element 6A via a conductive plug 22e buried in the interlayer insulating film 21. Then, the second electrode 23c of the resistance element 6A is electrically and mechanically connected to the relay wiring 25e.sub.1 formed in the first wiring layer. Then, although not illustrated in detail, the relay wiring 25e.sub.1 is electrically connected to the bit line BL.sub.1 (dendrite) formed in the second wiring layer on the interlayer insulating film 26 and extending in the Y direction via the conductive plug buried in the interlayer insulating film 26.

[0078] <Other Inverter Circuit>

[0079] As illustrated in FIG. 3, the respective gate electrodes 16p and 16d of the load FET Qp.sub.2 and the drive FET Qd.sub.2 configuring the other inverter circuit 4b are integrally molded and electrically connected to each other. That is, the respective gate electrodes 16p and 16d of the load FET Qp.sub.2 and the drive FET Qd.sub.2 are connected to each other to configure the input portion 4b.sub.1 (refer to FIG. 2).

[0080] As illustrated in FIG. 3, although not illustrated in detail, the first main electrode region 17p.sub.2 of the load FET Qp.sub.2 is electrically connected to the relay wiring 25a.sub.2 formed in the first wiring layer on the interlayer insulating film 21 via the conductive plug buried in the interlayer insulating film 21 on the semiconductor layer 10, similarly to that of the above-described load FET Qp.sub.1. On the other hand, although not illustrated in detail, the respective first main electrode regions 17d.sub.1 and 17t.sub.1 of the drive FET Qd.sub.2 and the transfer FET Qt.sub.2 are electrically connected to the relay wiring 25a.sub.2 via the conductive plug buried in the interlayer insulating film 21 on the semiconductor layer 10, similarly to those of the drive FET Qd.sub.1 and the transfer FET Qt.sub.1 described above. That is, the respective first main electrode regions 17p.sub.1, 17d.sub.1, and 17t.sub.1 of the load FET Qp.sub.2, the drive FET Qd.sub.2, and the transfer FET Qt.sub.2 are electrically connected to each other to configure the output portion 4b.sub.2 (refer to FIG. 2). The relay wiring 25a.sub.2 and the relay wiring 25c.sub.2 to 25e.sub.2 that will be described later are buried in the interlayer insulating film 24 on the interlayer insulating film 21.

[0081] As illustrated in FIG. 3, although not illustrated in detail, the second main electrode region 17d.sub.2 of the drive FET Qd.sub.2 is electrically connected to the relay wiring 25c.sub.2 formed in the first wiring layer via the conductive plug buried in the interlayer insulating film 21, similarly to that of the above-described drive FET Qd.sub.1. Then, the relay wiring 25c.sub.2 is electrically connected to the ground wiring 28c.sub.2 formed in the second wiring layer on the interlayer insulating film 26 and extending in the Y direction via the conductive plug buried in the interlayer insulating film 26 on the interlayer insulating film 24. The Vss potential (for example, 0 V) as the above-described first reference potential is applied to the ground wiring 28c.sub.2. That is, the Vss potential is supplied from the ground wire 28c.sub.2 to the second main electrode region 17d.sub.2 of the drive FET Qd.sub.2.

[0082] As illustrated in FIG. 3, although not illustrated in detail, the second main electrode region 17p.sub.2 of the load FET Qp.sub.2 is electrically connected to the relay wiring 25d.sub.2 formed in the first wiring layer via the conductive plug buried in the interlayer insulating film 21, similarly to that of the above-described load field effect transistor Qp.sub.1. Then, the relay wiring 25d.sub.2 is electrically connected to the power supply line 28d formed in the second wiring layer on the interlayer insulating film 26 and extending in the Y direction via the conductive plug buried in the interlayer insulating film 26. That is, the Vdd potential is supplied from the power supply line 28d to the second main electrode region 17p.sub.2 of the load FET Qp.sub.2.

[0083] As illustrated in FIG. 3, although not illustrated in detail, the second main electrode region 17t.sub.2 of the transfer FET Qt.sub.2 is electrically connected to the first electrode 23a of the resistance element 6B via the conductive plug buried in the interlayer insulating film 21, similarly to that of the above-described transfer FET Qt.sub.1. Then, the second electrode 23c of the resistance element 6B is electrically and mechanically connected to the relay wiring 25e.sub.2 formed in the first wiring layer. Then, the relay wiring 25e.sub.2 is electrically connected to the bit line bar BL.sub.2 (dendrite-) formed in the second wiring layer on the interlayer insulating film 26 and extending in the Y direction via the conductive plug buried in the interlayer insulating film 26.

[0084] <Two Inverter Circuits>

[0085] As illustrated in FIG. 3, although not illustrated in detail, the gate electrode 16p of the load FET Qp.sub.1 is electrically connected to the relay wiring 25a.sub.2 via the conductive plug buried in the interlayer insulating film 21. That is, the respective gate electrodes 16p and 16d of the load FET Qp.sub.1 and the drive FET Qd.sub.1 configuring the one inverter circuit 4a are electrically connected to the respective first main electrode regions 17p.sub.1 and 17d.sub.1 of the load FET Qp.sub.2 and the drive FET Qd.sub.2 configuring the other inverter circuit 4b and the first main electrode region 17t.sub.1 of the other transfer FET Qt.sub.2.

[0086] As illustrated in FIG. 3, although not illustrated in detail, the gate electrode 16p of the load FET Qp.sub.2 is electrically connected to the relay wiring 25a.sub.1 via the conductive plug buried in the interlayer insulating film 21. That is, the respective gate electrodes 16p and 16d of the load FET Qp.sub.2 and the drive FET Qd.sub.2 configuring the other inverter circuit 4b are electrically connected to the respective first main electrode regions 17p.sub.1 and 17d.sub.1 of the load FET Qp.sub.1 and the drive FET Qd.sub.1 configuring the one inverter circuit 4a and the first main electrode region 17t.sub.1 of the one transfer FET Qt.sub.1.

[0087] Note that, although not illustrated in FIG. 3, the respective gate electrodes 16p and 16d of the two transfer FETs Qt.sub.1 and Qt.sub.2 are electrically connected to the word line WL extending in the X direction. The word line WL is formed, for example, in a third wiring layer provided on the second wiring layer via the interlayer insulating film.

[0088] <<Write Operation and Product-Sum Calculation>>

[0089] Next, an operation of writing data to the memory cell 3 and a product-sum calculation will be described.

[0090] As a write operation to the memory cell 3, by applying the Vcc potential (for example, 1 V) to the word line WL to turn on the two transfer FETs Qt.sub.1 and Qt.sub.2, and setting the bit line BL.sub.1 to the Vdd potential and the bit line bar BL.sub.2 to the Vss potential, the output portion (storage node portion) 4a.sub.2 of one inverter circuit 4a is set to the Vcc potential and the output portion (storage node portion) 4b.sub.2 of the other inverter circuit 4b is set to the Vss potential. Even if the word line WL is set to the Vss potential, the flip-flop circuit 5 including the two inverter circuits 4a and 4b is stabilized.

[0091] On the other hand, at the time of a product-sum calculation (inference) using the memory cell array unit 2 as a product-sum calculation circuit, the bit line BL.sub.1 as a dendrite and the bit line bar BL.sub.2 as a dendrite bar (dendrite-) are precharged to the Vcc potential (for example, 1 V) in a state in which the flip-flop circuit 5 of the memory cell 3 stores data. Subsequently, a signal (for example, a pulse voltage) is sequentially input to the word line WL as an axon or to the plurality of word lines WL. When the signal reaches the Vcc potential, the respective transfer FETs Qt.sub.1 and Qt.sub.2 of the two inverter circuits 4a and 4b are turned on, and the drive FET Qd.sub.2 of the other inverter circuit 4b having the gate electrode 16d connected to the output portion (storage node portion) 4a.sub.2 of the one inverter circuit 4a is turned on. Therefore, the charge of the bit line bar BL.sub.2 is released toward the Vss potential, and the potential decreases. On the other hand, since the drive FET Qd.sub.1 of one inverter circuit 4a in which the gate electrode 16d is connected to the output portion 4b.sub.2 of the other inverter circuit 4b is in an OFF state, the charge of the bit line BL.sub.1 is not released, and the potential does not change.

[0092] Here, the potential of the bit line bar BL.sub.2 (dendrite-) changes with the CR time constant of the resistance value R of the resistance element 6B and the parasitic capacitance C parasitic on the bit line bar BL.sub.2. Therefore, by outputting or AD-converting a potential difference between the bit line BL.sub.1 (dendrite) and the bit line bar BL.sub.2 (dendrite-) in response to the CR time constant, the sum-product calculation with high power efficiency can be performed.

[0093] Note that the input potential to the word line WL, that is, the axon at the time of the sum-product calculation can be freely set separately from the Vcc potential or a write potential applied to the word WL at the time of writing data to the memory cell 3. In a case where the conductance of the load FETs Qp.sub.1 and Qp.sub.2 is greater than the conductance of the resistance elements 6A and 6B, writing of data to the memory cell 3 is performed by cutting off the Vcc potential or the Vss potential.

[0094] <<Main Effects of First Embodiment>>

[0095] As described above, the memory cell 3 of the semiconductor device 1 according to the first embodiment includes the flip-flop circuit 5, and the two transfer FETs Qt.sub.1 and Qt.sub.2, and further includes the two resistance elements 6A and 6B. Then, each of the two resistance elements 6A and 6B has a resistance value of 1 MΩ or more, which is necessary in a case where a bit line as a summation line is charged with product-sum charge for multiple bits (for example, 1024), during the product-sum calculation process, and is disposed within an occupied area of the memory cell 3. Therefore, according to the semiconductor device 1 of the first embodiment, it is possible to perform a product-sum calculation with high power efficiency while maintaining a small area of the memory cell 3.

Second Embodiment

[0096] In a second embodiment, an SRAM-type memory cell including ten field effect transistors and two resistance elements will be described.

[0097] <<Configuration of Memory Cell Array Unit>>

[0098] A semiconductor device 1A according to the second embodiment includes a memory cell array unit 2A illustrated in FIG. 5. As illustrated in FIG. 5, in the memory cell array unit 2A, a plurality of memory cells 3A are arranged in a matrix on a two-dimensional plane including the X direction and the Y direction. Furthermore, in the memory cell array unit 2A, word lines WL extending in the X direction are arranged for each of memory cells 3A arranged in the Y direction. Furthermore, in the memory cell array unit 2A, complementary bit lines (a bit line BL.sub.1 and a bit line bar BL.sub.2 (BL-)) extending in the Y direction are arranged for each of the memory cells 3A arranged in the X direction. Furthermore, in the memory cell array unit 2A, complementary dendrite lines (dendrite line DL.sub.1 and dendrite line bar (dendrite line-) DL.sub.2) extending in the Y direction are arranged for each of the memory cells 3A arranged in the X direction. Then, each memory cell 3A of the plurality of memory cells 3A is disposed at an intersection between the corresponding word line WL and the complementary bit lines (BL.sub.1 and BL.sub.2) and complementary dendrite lines (DL.sub.1 and DL.sub.2).

[0099] Here, during a product-sum calculation process (inference in a neural network), the memory cell array unit 2A functions as a product-sum calculation circuit, the word line WL functions as an axon, and the dendrite line DL.sub.1 and the dendrite line bar DL.sub.2 function as a dendrite and a dendrite bar (dendrite-).

[0100] <<Configuration of Memory Cell>>As illustrated in FIG. 6, each memory cell 3A of the plurality of memory cells 3A includes a flip-flop circuit 5, two transfer FETs Qt.sub.1 and Qt.sub.2 as first transfer field effect transistors, two transfer FETs Qt.sub.3 and Qt.sub.4 as second transfer field effect transistors, two drive FETs Qd.sub.3 and Qd.sub.4 as second drive field effect transistors, and two resistance elements 6A and 6B.

[0101] The flip-flop circuit 5 includes two inverter circuits 4a and 4b, and has a configuration in which input portions 4a.sub.1 and 4b.sub.1 and the output portions (storage node portions) 4a.sub.2 and 4b.sub.2 of two inverter circuits 4a and 4b are alternately cross-joined.

[0102] One inverter circuit 4a of the two inverter circuits 4a and 4b has a configuration in which a load FET (pull-up transistor) Qp.sub.1 and a drive FET (pull-down transistor) Qd.sub.1 as a first drive field effect transistor are connected in series. Furthermore, the other inverter circuit 4b has a configuration in which a load FET Qp.sub.2 and a drive FET Qd.sub.2 as a first drive field effect transistor are connected in series.

[0103] Similarly to the two drive FETs Qd.sub.1 and Qd.sub.2 and the two transfer FETs Qt.sub.1 and Qt.sub.2, the two drive FETs Qd.sub.3 and Qd.sub.4 and the two transfer FETs Qt.sub.3 and Qt.sub.4 include a gate insulating film, a gate electrode (control electrode), and a pair of first main electrode region and second main electrode region functioning as a source region and a drain region, and electrical conduction between the first main electrode region and the second main electrode region is controlled by a gate signal input to the gate electrode. Then, the FETs Qd.sub.3, Qd.sub.4, Qt.sub.3, and Qt.sub.4 are also configured by n-channel conductivity type MOSFETs, for example.

[0104] As illustrated in FIG. 6, in one transfer FET Qt.sub.1 of the two transfer FETs Qt.sub.1 and Qt.sub.2, similarly to the transfer FET Qt.sub.1 of the first embodiment described above, the gate electrode is electrically connected to the word line WL, and the first main electrode region of the pair of first and second main electrode regions is electrically connected to the output portion 4a.sub.2 of one inverter circuit 4a. Then, unlike the transfer FET Qt.sub.1 of the first embodiment described above, the second main electrode region of one transfer FET Qt.sub.1 is electrically connected to the bit line BL.sub.1 without passing through the resistance element 6A.

[0105] As illustrated in FIG. 6, in the other transfer FET Qt.sub.2 of the two transfer FETs Qt.sub.1 and Qt.sub.2, similarly to the transfer FET Qt.sub.2 of the first embodiment described above, the gate electrode is electrically connected to the word line WL, and the first main electrode region of the pair of first and second main electrode regions is electrically connected to the output portion 4b.sub.2 of the other inverter circuit 4b. Then, in the other transfer FET Qt.sub.2, unlike the transfer FET Qt.sub.2 of the first embodiment described above, the second main electrode region is electrically connected to the bit line bar BL.sub.2without passing through the resistance element 6B.

[0106] As illustrated in FIG. 6, one drive FET Qd.sub.3 of the two drive FETs Qd.sub.3 and Qd.sub.4 has a gate electrode electrically connected to the gate electrode of each of the load FET Qp.sub.1 and the drive FET Qd.sub.1 configuring one inverter circuit 4a. That is, the gate electrode of the one drive FET Qd.sub.3 is electrically connected to the input portion 4a.sub.1 of the one inverter circuit 4a.

[0107] As illustrated in FIG. 6, the other drive FET Qd.sub.4 of the two drive FETs Qd.sub.3 and Qd.sub.4 has a gate electrode electrically connected to the gate electrode of each of the load FET Qp.sub.2 and the drive FET Qd.sub.2 configuring the other inverter circuit 4b. That is, the gate electrode of the other drive FET Qd.sub.4 is electrically connected to the input portion 4b.sub.1 of the other inverter circuit 4b.

[0108] As illustrated in FIG. 6, the first main electrode region of each of the two drive FETs Qd3 and Qd4 is electrically connected to the ground wiring to which the Vss potential as a first reference potential is applied.

[0109] As illustrated in FIG. 6, in one transfer FET Qt.sub.3 of the two transfer FETs Qt.sub.3 and Qt.sub.4, the gate electrode is electrically connected to the word line WL, and the first main electrode region of the pair of first and second main electrode regions is electrically connected to the second main electrode region of the drive FET Qd.sub.3. Then, in the one transfer FET Qt.sub.3, the second main electrode region is electrically connected to the first electrode on one end side of the resistance element 6A. Then, the second main electrode on the other end side of the resistance element 6A is electrically connected to the dendrite line DL.sub.1.

[0110] As illustrated in FIG. 6, in the other transfer FET Qt.sub.4 of the two transfer FETs Qt.sub.3 and Qt.sub.4, the gate electrode is electrically connected to the word line WL, and the first main electrode region of the pair of first and second main electrode regions is electrically connected to the second main electrode region of the drive FET Qd.sub.4. Then, in the other transfer FET Qt.sub.4, the second main electrode region is electrically connected to the first electrode on one end side of the resistance element 6B. Then, the second main electrode on the other end side of the resistance element 6B is electrically connected to the dendrite line bar (dendrite line-) DL.sub.2.

[0111] Although not illustrated, the dendrite line DL.sub.1 and the dendrite line bar DL.sub.2 are formed in the second wiring layer and extend in the Y direction, for example, similarly to the bit line BL.sub.1 and the bit line bar BL.sub.2.

[0112] The two resistance elements 6A and 6B of the second embodiment also have an MIM structure similarly to the resistance elements 6A and 6B of the first embodiment described above, can obtain a desired resistance value with a small area, and are arranged within an occupied area of the memory cell. Then, a resistance value of each of the two resistance elements 6A and 6B is preferably larger than a channel resistance value of the transfer FETs Qt.sub.3 and Qt.sub.4, and more preferably 1 MΩ or more.

[0113] <<Write Operation and Product-Sum Calculation>>

[0114] Next, an operation of writing data to the memory cell 3A and a product-sum calculation will be described.

[0115] As a write operation to the memory cell 3A, by applying the Vcc potential (for example, 1 V) to the word line WL to turn on the two transfer FETs Qt.sub.1 and Qt.sub.2 and setting the bit line BL.sub.1 to the Vcc potential and the bit line bar BL.sub.2 to the Vss potential, the output portion (storage node portion) 4a.sub.2 of one inverter circuit 4a is set to the Vcc potential and the output portion (storage node portion) 4b.sub.2 of the other inverter circuit 4b is set to the Vss potential. Even if the word line WL is set to the Vss potential, the flip-flop circuit 5 including the two inverter circuits 4a and 4b is stabilized.

[0116] On the other hand, at the time of a product-sum calculation (inference) using the memory cell array unit 2A as a product-sum calculation circuit, the dendrite line DL.sub.1 and the dendrite line bar DL.sub.2 are precharged to the Vcc potential (for example, 1 V) in a state in which the flip-flop circuit 5 of the memory cell 3A stores data. Subsequently, a signal (for example, a pulse voltage) is sequentially input to the word line WL as an axon or to the plurality of word lines WL. When the signal reaches the Vcc potential, the two transfer FETs Qt3 and Qt4 are turned on, and the other drive FET Qd.sub.4 having the gate electrode connected to the output portion (storage node portion) 4a.sub.2 of the one inverter circuit 4a is turned on. Therefore, the charge of the dendrite line bar DL.sub.2 is released toward the Vss potential, and the potential decreases. On the other hand, since one drive FET Qd.sub.3 having the gate electrode connected to the output portion 4b.sub.2 of the other inverter circuit 4b is in an OFF state, the charge of the dendrite line DL.sub.1 is not released, and the potential does not change.

[0117] Here, the potential of the dendrite line bar DL.sub.2 (dendrite line-) changes with the CR time constant of the resistance value R of the resistance element 6B and the parasitic capacitance C parasitic on the dendrite line bar DL.sub.2. Therefore, by outputting or AD-converting a potential difference between the dendrite line DL.sub.1 and the dendrite line bar DL.sub.2 (dendrite line-) in response to the CR time constant, the sum-product calculation with high power efficiency can be performed.

[0118] Note that the input potential to the word line WL, that is, the axon at the time of the sum-product calculation can be freely set separately from the Vcc potential or a write potential applied to the word WL at the time of writing data to the memory cell 3A. Writing of data into the flip-flop circuit 5 of the memory cell 3A is usually performed at a high speed through an SRAM operation. The current consumption can be reduced by reducing the Vcc potential during the SRAM operation and increasing a threshold voltage Vth of each of the load FETs Qp.sub.1 and Qp.sub.2 and the drive FTEs Qd.sub.1 and Qd.sub.2.

[0119] <<Main Effects of Second Embodiment>>

[0120] As described above, the memory cell 3A of the semiconductor device 1A according to the second embodiment includes the flip-flop circuit 5, the four transfer FETs Qt.sub.1, Qt.sub.2, Qt.sub.3, and Qt.sub.4, and the two drive FETs Qd.sub.3 and Qd.sub.4, and further includes the two resistance elements 6A and 6B. Then, each of the two resistance elements 6A and 6B has a resistance value of 1 MΩ or more, which is necessary in a case where a bit line as a summation line is charged with product-sum charge for multiple bits (for example, 1024), during the product-sum calculation process, and is disposed within an occupied area of the memory cell 3A. Therefore, according to the semiconductor device 1A of the first embodiment, similarly to the semiconductor device 1 of the first embodiment described above, it is possible to perform a product-sum calculation with high power efficiency while maintaining a small area of the memory cell 3A.

Third Embodiment

[0121] In a third embodiment, an SRAM-type memory cell including nine field effect transistors and two resistance elements will be described.

[0122] <<Configuration of Memory Cell Array Unit>>

[0123] The semiconductor device 1B according to the third embodiment includes a memory cell array unit 2B illustrated in FIG. 7. As illustrated in FIG. 7, in the memory cell array unit 2B, a plurality of memory cells 3B is arranged in a matrix on a two-dimensional plane including the X direction and the Y direction. Furthermore, in the memory cell array unit 2B, a word line WL and an axon line AL extending in the X direction are arranged for each of memory cells 3B arranged in the Y direction. Furthermore, in the memory cell array unit 2B, bit line bars BL.sub.2 (BL-) extending in the Y direction are arranged for the memory cells 3B arranged in the X direction. In the memory cell array unit 2B, complementary dendrite lines (a dendrite line DL.sub.1 and a dendrite line bar (dendrite line-) DL.sub.2) extending in the Y direction are arranged for each of the memory cells 3B arranged in the X direction. Then, each memory cell 3B of the plurality of memory cells 3B is disposed at an intersection between the corresponding word line WL, the complementary bit lines (BL.sub.1 and BL.sub.2), and the dendrite line DL.sub.2.

[0124] In the memory cell array unit 2B, the bit line BL.sub.1 of the second embodiment described above is eliminated, and the dendrite line DL.sub.1 is used as a bit line when writing data to the memory cell 3B.

[0125] Here, during a product-sum calculation process (inference in a neural network), the memory cell array unit 2B functions as a product-sum calculation circuit, the axon line AL functions as an axon, and the dendrite line DL.sub.1 and the dendrite line bar DL.sub.2 function as a dendrite and a dendrite bar (dendrite-). On the other hand, when data is written to the memory cell 3B, the dendrite line DL.sub.1 functions as a bit line.

[0126] <<Configuration of Memory Cell>>

[0127] As illustrated in FIG. 8, each memory cell 3B of the plurality of memory cells 3B includes a flip-flop circuit 5, a transfer FET Qt.sub.2 as a first transfer field effect transistor, two transfer FETs Qt.sub.3 and Qt.sub.4 as second transfer field effect transistors, two drive FETs Qd.sub.3 and Qd.sub.4 as second drive field effect transistors, and two resistance elements 6A and 6B. The memory cell 3B has a configuration basically similar to that of the memory cell 3A of the second embodiment described above, and is different in that one transfer FET Qt.sub.1 is eliminated and the gate electrode of each of the two transfer FETs Qt.sub.3 and Qt.sub.4 is electrically connected to the axon line AL. Other configurations are similar to those of the memory cell 3A of the second embodiment described above. However, similarly to the second embodiment, the transfer FET Qt.sub.2 may be added.

[0128] The two resistance elements 6A and 6B of the third embodiment also have an MIM structure similarly to the resistance elements 6A and 6B of the first embodiment described above, can obtain a desired resistance value with a small area, and are arranged within an occupied area of the memory cell. Then, a resistance value of each of the two resistance elements 6A and 6B is preferably larger than a channel resistance value of the transfer FETs Qt.sub.3 and Qt.sub.4, and more preferably 1 MΩ or more.

[0129] <Write Operation and Product-Sum Calculation>

[0130] Next, an operation of writing data to the memory cell 3B and a product-sum calculation will be described.

[0131] As the write operation of data to the memory cell 3B, by applying the Vcc potential (for example, 1 V) to the word line WL to turn on the transfer FET Qt.sub.2 and setting the bit line bar BL.sub.2 to the Vss potential, the output portion (storage node portion) 4a.sub.2 of one inverter circuit 4a is set to the Vcc potential, the output portion (storage node portion) 4b.sub.2 of the other inverter circuit 4b is set to the Vss potential. Even if the word line WL is set to the Vss potential, the flip-flop circuit 5 including the two inverter circuits 4a and 4b is stabilized. By setting the bit line bar BL.sub.2 to the Vcc potential, the output portion (storage node portion) 4a.sub.2 of one inverter circuit 4a is set to the Vss potential, the output portion (storage node portion) 4b.sub.2 of the other inverter circuit 4b is set to the Vcc potential. Even if the word line WL is set to the Vss potential, the flip-flop circuit 5 including the two inverter circuits 4a and 4b is stabilized.

[0132] On the other hand, during a product-sum calculation (inference) using the memory cell array unit 2B as a product-sum calculation circuit, the dendrite line DL.sub.1 and the dendrite line bar DL.sub.2 are precharged to the Vcc potential (for example, 1 V) in a state in which the flip-flop circuit 5 of the memory cell 3B stores data. Subsequently, a signal (for example, a pulse voltage) is input sequentially to the axon line AL as an axon or to the plurality of axon lines AL. When the signal reaches the Vcc potential, the two transfer FETs Qt3 and Qt4 are turned on, and the other drive FET Qd.sub.4 having the gate electrode 16d connected to the output portion (storage node portion) 4a.sub.2 of the one inverter circuit 4a is turned on. Therefore, the charge of the dendrite line bar DL.sub.2 is released toward the Vss potential, and the potential decreases. On the other hand, since one drive FET Qd.sub.3 having the gate electrode 16d connected to the output portion 4b2 of the other inverter circuit 4b is in an OFF state, the charge of the dendrite line DL.sub.1 is not released, and the potential does not change.

[0133] Here, the potential of the dendrite line bar DL.sub.2 (dendrite line-) changes with the CR time constant of the resistance value R of the resistance element 6B and the parasitic capacitance C parasitic on the dendrite line bar DL.sub.2. The resistance elements 6A and 6B are connected to the drive FETs Qd3 and Qd4 for product-sum calculation, and thus do not affect writing of data to the memory cell 3B. Therefore, by outputting or AD-converting a potential difference between the dendrite line DL.sub.1 and the dendrite line bar DL.sub.2 (dendrite line-) in response to the CR time constant, the sum-product calculation with high power efficiency can be performed.

[0134] In the memory cell 3B of the third embodiment, since the word line WL is in an OFF state in which the potential is not supplied even in an ON state in which the Vcc potential is supplied to the axon line AL, the transfer FET Qt.sub.2 of the memory cell can be turned off even during the product-sum calculation, and the low power consumption and the stabilization of the memory operation can be achieved.

[0135] Note that the input potential to the word line WL, that is, the axon at the time of the sum-product calculation can be freely set separately from the Vcc potential or a write potential applied to the word WL at the time of writing data to the memory cell 3A. Writing of data into the flip-flop circuit 5 of the memory cell 3A is usually performed at a high speed through an SRAM operation. By reducing the Vcc potential during the SRAM operation and increasing the threshold voltage Vth of each of the transfer FETs Qp.sub.1 and Qp.sub.2 and the drive FTEs Qd.sub.1 and Qd.sub.2, the current consumption can be reduced.

[0136] <<Main Effects of Third Embodiment>>

[0137] As described above, the memory cell 3B of the semiconductor device 1B according to the third embodiment includes the flip-flop circuit 5, the three transfer FETs Qt.sub.2, Qt.sub.3, and Qt.sub.4, the two drive FETs Qd.sub.3 and Qd.sub.4, and further includes the two resistance elements 6A and 6B. Then, each of the two resistance elements 6A and 6B has a resistance value of 1 MΩ or more, which is necessary in a case where a bit line as a summation line is charged with product-sum charge for multiple bits (for example, 1024), during a product-sum calculation process, and is disposed within an occupied area of the memory cell 3B. Therefore, also in the semiconductor device 1B according to the third embodiment, similarly to the semiconductor device 1 according to the first embodiment described above, it is possible to perform a product-sum calculation with high power efficiency while maintaining a small area of the memory cell 3B.

Fourth Embodiment

[0138] In a fourth embodiment, an SRAM-type memory cell including six field effect transistors and four tunnel field effect transistors will be described.

[0139] <<Configuration of Memory Cell Array Unit>>

[0140] A semiconductor device 1C according to the second embodiment includes a memory cell array unit 2C illustrated in FIG. 9. As illustrated in FIG. 9, in the memory cell array unit 2C, a plurality of memory cells 3C is arranged in a matrix on a two-dimensional plane including the X direction and the Y direction. Furthermore, in the memory cell array unit 2A, similarly to the memory cell array unit 2A of the second embodiment described above, word lines WL extending in the X direction are arranged for each of memory cells 3C arranged in the Y direction, complementary bit lines (a bit line BL.sub.1 and a bit line bar BL.sub.2 (BL.sub.2-)) extending in the Y direction are arranged for each of the memory cells 3C arranged in the X direction, and complementary dendrite lines (dendrite line DL.sub.1 and dendrite line bar (dendrite line-) DL.sub.2) extending in the Y direction are arranged for each of the memory cells 3C arranged in the X direction. Then, each memory cell 3B of the plurality of memory cells 3B is disposed at an intersection between the corresponding word line WL and the complementary bit lines (BL.sub.1 and BL.sub.2) and complementary dendrite lines (DL.sub.1 and DL.sub.2).

[0141] Here, during a product-sum calculation process (inference in the neural network), the memory cell array unit 2C functions as a product-sum calculation circuit, the axon line AL illustrated in FIG. 10 functions as an axon, and the dendrite line DL.sub.1 and the dendrite line bar DL.sub.2 function as a dendrite and a dendrite bar (dendrite-).

[0142] <<Configuration of Memory Cell>>

[0143] As illustrated in FIG. 10, each memory cell 3C of the plurality of memory cells 3C includes a flip-flop circuit 5, two transfer FETs Qt.sub.1 and Qt.sub.2 as first transfer field effect transistors, two transfer tunnel FETs Qt.sub.5 and Qt.sub.6 as second transfer tunnel field effect transistors, and two drive tunnel FETs Qd.sub.5 and Qd.sub.6 as second drive tunnel field effect transistors. The memory cell 3C has a configuration basically similar to that of the memory cell 3A of the second embodiment described above, and includes drive tunnel FETs Qd.sub.5 and Qd.sub.6 and transfer tunnel FETs Qt.sub.5 and Qt.sub.6 instead of the resistance elements 6A and 6B, the drive FTEs Qd.sub.3 and Qd.sub.4, and the transfer FETs Qt.sub.3 and Qt.sub.4 of the second embodiment described above.

[0144] The two drive tunnel FETs Q.sub.5 and Qd.sub.6 each have a gate electrode (control electrode) and a pair of n-type first main electrode region and p-type second main electrode region functioning as a source region and a drain region, and electrical conduction between the n-type first main electrode region and the p-type main electrode separation region is controlled by a gate signal input to the gate electrode.

[0145] As illustrated in FIG. 10, the gate electrode of one drive tunnel FET Q.sub.5 of the two drive tunnel FETs Qd.sub.5 and Qd.sub.6 is electrically connected to the gate electrode of each of the load FET Qp.sub.1 and the drive FET Qd.sub.5 configuring one inverter circuit 4a. That is, the gate electrode of the one drive tunnel FET Qd.sub.5 is electrically connected to the input portion 4a.sub.1 of the one inverter circuit 4a.

[0146] As illustrated in FIG. 10, in the other drive tunnel FET Qd.sub.6 of the two drive tunnel FETs Qd.sub.5 and Qd.sub.6, the gate electrode is electrically connected to the gate electrode of each of the load FET Qp.sub.2 and the drive FET Qd.sub.2 configuring the other inverter circuit 4b. That is, the gate electrode of the other drive tunnel FET Qd.sub.6 is electrically connected to the input portion 4b.sub.1 of the other inverter circuit 4b.

[0147] As illustrated in FIG. 10, the n-type first main electrode region of each of the two drive tunnel FETs Qd.sub.5 and Qd.sub.6 is electrically connected to the axon line AL.

[0148] As illustrated in FIG. 10, in one transfer FET Qt.sub.5 of the two transfer tunnel FETs Qt.sub.5 and Qt.sub.6, the gate electrode is electrically connected to the word line WL, and the p-type second main electrode region is electrically connected to the p-type second main electrode region of one drive tunnel FET Qd.sub.5. Then, in the one transfer tunnel FET Qt.sub.5, the n-type first main electrode region is electrically connected to the dendrite line DL.sub.1.

[0149] As illustrated in FIG. 10, in the other transfer tunnel FET Qt.sub.6 of the two transfer tunnel FETs Qt.sub.5 and Qt.sub.6, the gate electrode is electrically connected to the word line WL, and the p-type second main electrode region is electrically connected to the p-type second main electrode region of the other drive tunnel FET Qd.sub.6. Then, in the other transfer tunnel FET Qt.sub.6, the n-type first main electrode region is electrically connected to the dendrite line bar (dendrite line) DL.sub.2.

[0150] As illustrated in FIG. 10, in the two drive tunnel FETs Qd.sub.5 and Qd.sub.6 and the two transfer tunnel FETs Qt.sub.5 and Qt.sub.6, a pair of main electrode regions functioning as a source region and a drain region includes the n-type first main electrode region and the p-type second main electrode region. Therefore, a pn-type parasitic diode PD.sub.1 is formed in an equivalent circuit on a conductive path connecting the one transfer tunnel FET Qt.sub.5 to the dendrite line DL1. Furthermore, a pn-type parasitic diode PD.sub.2 is formed in an equivalent circuit on a conductive path connecting the other transfer tunnel FET Qt.sub.6 to the dendrite line bar DL.sub.2.

[0151] <<Write Operation and Product-Sum Calculation>>

[0152] Next, an operation of writing data to the memory cell 3C and a product-sum calculation will be described.

[0153] As a write operation to the memory cell 3C, by applying the Vcc potential (for example, 1 V) to the word line WL to turn on the two transfer FETs Qt.sub.1 and Qt.sub.2 and setting the bit line BL.sub.1 to the Vcc potential and the bit line bar BL.sub.2 to the Vss potential, the output portion (storage node portion) 4a.sub.2 of one inverter circuit 4a is set to the Vcc potential and the output portion (storage node portion) 4b.sub.2 of the other inverter circuit 4b is set to the Vss potential. Even if the word line WL is set to the Vss potential, the flip-flop circuit 5 including the two inverter circuits 4a and 4b is stabilized.

[0154] On the other hand, at the time of a product-sum calculation (inference) using the memory cell array unit 2C as a product-sum calculation circuit, the dendrite line DL.sub.1 and the dendrite line bar DL.sub.2 are precharged to the Vss potential (for example, 0 V) in a state in which the flip-flop circuit 5 of the memory cell 3C stores data. Subsequently, a signal (for example, a pulse voltage) is input to the axon line AL. When the signal reaches the Vcc potential, since the other drive tunnel FET Qd.sub.6 having the gate electrode connected to the output portion (storage node portion) 4a.sub.2 of the one inverter circuit 4a is in an ON state, the dendrite line bar DL.sub.2 is charged by being supplied with electric charge from the axon line AL. In this case, even when the word line WL has the Vss potential and the other transfer tunnel FET Qt6 is in an OFF state, since the source region (second main electrode region) side of the transfer tunnel FET Qt6 has a high potential, the parasitic diode PD2 operates in the forward direction, and the charging of the dendrite line bar DL.sub.2 is not inhibited.

[0155] On the other hand, since one drive tunnel FET Qd.sub.5 electrically connected to the output portion (storage node portion) 4b.sub.2 of the other inverter circuit 4b is in an OFF state, the dendrite line DL.sub.1 is not charged with the electric charge from the axon line AL and the potential is maintained.

[0156] The potential of the dendrite line bar DL.sub.2 changes with the CR time constant of the channel resistance value R of each of the drive tunnel FET Qd6 and the transfer tunnel FET Qt6 and the parasitic capacitance C of the dendrite line bar DL.sub.2. The transfer FET connected to the storage node portion of the flip-flop circuit is a normal MOSFET, and thus does not affect writing of data to the memory cell. Furthermore, even if the signal of the axon line AL becomes the Vss potential (ground potential) and the dendrite line bar DL.sub.2 has a high potential, the transfer tunnel FET Qt6 is in an OFF state, and the parasitic diode PD2 is in an opposite direction, so that the electric charge does not flow back from the dendrite line bar DL.sub.2 to the axon line AL. Therefore, by outputting or AD-converting a potential difference between the dendrite line DL.sub.1 and the dendrite line bar DL.sub.2 (dendrite line-) in response to the CR time constant, the sum-product calculation with high power efficiency can be performed.

[0157] The two drive tunnel FETs Q.sub.5 and Qd.sub.6 and the two transfer tunnel FETs Qt.sub.5 and Qt.sub.6 can increase the channel resistance value with a small area compared with a normal MOSFET. That is, during the product-sum calculation process, it is possible to obtain a resistance value of 1 MΩ or more, which is necessary in a case where a bit line as a summation line is charged with product-sum charge for multiple bits (for example, 1024) with a small area compared with the MOSFET. Therefore, also in the semiconductor device 1A according to the fourth embodiment, similarly to the semiconductor device 1 according to the first embodiment described above, it is possible to perform a product-sum calculation with high power efficiency while maintaining a small area of the memory cell 3C.

[0158] Note that a channel resistance value of the two drive tunnel FETs Q.sub.5 and Qd.sub.6 and the two transfer tunnel FETs Qt.sub.5 and Qt.sub.6 is preferably larger than a channel resistance value of the transfer FETs Qt.sub.3 and Qt.sub.4, and more preferably 1 MΩ or more.

MODIFICATION EXAMPLES

First Modification Example

[0159] FIG. 11 is an equivalent circuit diagram of a memory cell according to a first modification example of the fourth embodiment.

[0160] As illustrated in FIG. 11, a memory cell 3C.sub.1 basically has a configuration similar to that of the memory cell of the fourth embodiment described above, and has a configuration in which the two transfer tunnel FETs Qt.sub.5 and Qt.sub.6 illustrated in FIG. 10 are omitted compared with the memory cell of the fourth embodiment.

[0161] That is, as illustrated in FIG. 11, the memory cell 3C.sub.1 of the first modification example includes the flip-flop circuit 5, the two transfer FETs Qt.sub.1 and Qt.sub.2, and the two drive tunnel FETs Q.sub.5 and Qd.sub.6, and does not include the two transfer tunnel FETs Qt.sub.5 and Qt.sub.6 illustrated in FIG. 10. Therefore, the p-type second main electrode region of one drive tunnel FET Q.sub.5 is electrically connected to the dendrite line DL.sub.1 without passing through the transfer tunnel FET, and the p-type second main electrode region of the other drive tunnel FET Qd.sub.6 is electrically connected to the dendrite line bar DL.sub.2 without passing through the transfer tunnel FET.

[0162] Also in the memory cell 3C.sub.1 of the first modification example configured as described above, effects similar to those of the above-described fourth embodiment can be achieved.

Second Modification Example

[0163] FIG. 12 is an equivalent circuit diagram of a memory cell according to a second modification example of the fourth embodiment.

[0164] As illustrated in FIG. 12, a memory cell 3C.sub.2 basically has a configuration similar to that of the memory cell 3C of the above-described fourth embodiment, and has a configuration in which the drive tunnel FET Qd.sub.6 and the transfer tunnel FET Qt.sub.6 on the other inverter circuit 4b side illustrated in FIG. 10 are omitted compared with the memory cell 3C of the fourth embodiment.

[0165] That is, as illustrated in FIG. 12, the memory cell 3C.sub.2 of the second modification example includes the flip-flop circuit 5, the two transfer FETs Qt.sub.1 and Qt.sub.2, and further includes one drive tunnel FET Q.sub.5 and one transfer tunnel FET Qt.sub.6. Then, in the memory cell array unit, the dendrite line bar DL.sub.2 illustrated in FIG. 10 is omitted.

[0166] Also in the memory cell 3C.sub.2 of the second modification example configured as described above, effects similar to those of the above-described fourth embodiment can be achieved.

[0167] Note that the present technology may have the following configurations.

[0168] (1)

[0169] A semiconductor device including:

[0170] a memory cell array in which a plurality of memory cells is arranged in a matrix, in which

[0171] each memory cell of the plurality of memory cells includes

[0172] a flip-flop circuit including two inverter circuits in each of which a load field effect transistor and a drive field effect transistor are connected in series, input portions and output portions of the two inverter circuits being cross-joined to each other,

[0173] two transfer field effect transistors each having a gate electrode connected to a word line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the output portions of the two inverter circuits, and

[0174] two resistance elements of which one end sides are respectively connected to the second main electrode regions of the two transfer field effect transistors and other end sides are respectively connected to a bit line and a bit line bar.

[0175] (2)

[0176] The semiconductor device according to (1), in which a resistance value of the resistance elements is greater than a channel resistance value of the transfer field effect transistors.

[0177] (3)

[0178] The semiconductor device according to (1), in which a resistance value of the resistance elements is 1 MΩ or more.

[0179] (4)

[0180] The semiconductor device according to (1) or (2), in which when a product-sum calculation is performed, the word line functions as an axon, the bit line functions as a dendrite, and the bit line bar functions as a dendrite bar.

[0181] (5)

[0182] A semiconductor device including:

[0183] a memory cell array in which a plurality of memory cells is arranged in a matrix, in which

[0184] each memory cell of the plurality of memory cells includes

[0185] a flip-flop circuit including two inverter circuits in each of which a load field effect transistor and a first drive field effect transistor are connected in series, input portions and output portions of the two inverter circuits being cross-joined to each other,

[0186] two first transfer field effect transistors each having a gate electrode connected to a word line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the output portions of the two inverter circuits, and the second main electrode regions being respectively connected to a bit line and a bit line bar,

[0187] two second drive field effect transistors each having a gate electrode and a pair of first and second main electrode regions, the gate electrodes being respectively connected to the input portions of the two inverter circuits, and the first main electrode regions being connected to each other,

[0188] two second transfer field effect transistors each having a gate electrode connected to the word line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the second main electrode regions of the two second drive field effect transistors, and

[0189] two resistance elements of which one end sides are respectively connected to the second main electrode regions of the two second transfer field effect transistors and other end sides are respectively connected to a dendrite line and a dendrite line bar.

[0190] (6)

[0191] The semiconductor device according to (5), in which a resistance value of the resistance elements is greater than a channel resistance value of the transfer field effect transistors.

[0192] (7)

[0193] The semiconductor device according to (5), in which a resistance value of the resistance elements is 1 MΩ or more.

[0194] (8)

[0195] The semiconductor device according to any one of (5) to (7), in which when a product-sum calculation is performed, the word line functions as an axon.

[0196] (9)

[0197] A semiconductor device including:

[0198] a memory cell array in which a plurality of memory cells is arranged in a matrix, in which

[0199] each memory cell of the plurality of memory cells includes

[0200] a flip-flop circuit including two inverter circuits in each of which a load field effect transistor and a first drive field effect transistor are connected in series, input portions and output portions of the two inverter circuits being cross-joined to each other,

[0201] a first transfer field effect transistor having a gate electrode connected to a word line, and a pair of first and second main electrode regions, the first main electrode region being connected to the output portion of the other inverter circuit of the two inverter circuits, and the second main electrode region being connected to a bit line bar,

[0202] two second drive field effect transistors each having a gate electrode and a pair of first and second main electrode regions, the gate electrodes being respectively connected to the input portions of the two inverter circuits, and the first main electrode regions being connected to each other,

[0203] two second transfer field effect transistors each having a gate electrode connected to an axon line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the second main electrode regions of the two second drive field effect transistors, and

[0204] two resistance elements of which one end sides are respectively connected to the second main electrode regions of the two second transfer field effect transistors and other end sides are respectively connected to a dendrite line and a dendrite line bar.

[0205] (10)

[0206] The semiconductor device according to (9), in which a resistance value of the resistance elements is greater than a channel resistance value of the transfer field effect transistors.

[0207] (11)

[0208] The semiconductor device according to (9), in which a resistance value of the resistance elements is 1 MΩ or more.

[0209] (12)

[0210] A semiconductor device including:

[0211] a memory cell array in which a plurality of memory cells is arranged in a matrix, in which

[0212] each memory cell of the plurality of memory cells includes

[0213] a flip-flop circuit including two inverter circuits in which a load field effect transistor and a first drive field effect transistor are connected in series, input portions and output portions of the two inverter circuits being cross-joined to each other,

[0214] two first transfer field effect transistors each having a gate electrode connected to a word line, and a pair of first and second main electrode regions, the first main electrode regions being respectively connected to the output portions of the two inverter circuits, and the second main electrode regions being respectively connected to a bit line and a bit line bar,

[0215] two second drive tunnel field effect transistors each having a gate electrode and a pair of n-type first main electrode region and p-type second main electrode region, the gate electrodes being respectively connected to the input portions of the two inverter circuits, and the n-type first main electrode region being connected to an axon line, and

[0216] two second transfer tunnel field effect transistors each having a gate electrode connected to the word line, and a pair of n-type first main electrode region and p-type second main electrode region, the p-type second main electrode regions being respectively connected to the p-type second main electrode regions of the two second drive tunnel field effect transistors, and the n-type first main electrode regions being respectively connected to a dendrite line and a dendrite line bar.

[0217] The scope of the present technology is not limited to the illustrated and described exemplary embodiments, but also includes all embodiments that provide equivalent effects to those for which the present technology is intended. Moreover, the scope of the present technology is not limited to the combinations of the features of the invention defined by the claims, but may be defined by any desired combination of specific features among all the disclosed respective features.

REFERENCE SIGNS LIST

[0218] 1 Semiconductor device [0219] 2 Memory cell array unit [0220] 3 Memory cell [0221] 4a, 4b Inverter circuit [0222] 4a.sub.1, 4b.sub.1 Input portion [0223] 4a.sub.2, 4b.sub.2 Output portion (storage node portion) [0224] 5 Flip-flop circuit [0225] 6A, 6B Resistance element [0226] Qd.sub.1, Qd.sub.2, Qd.sub.3, Qd.sub.4 Drive field effect transistor (pull-down transistor) [0227] Qp.sub.1, Qp.sub.2 Load field effect transistor (pull-up transistor) [0228] Qt.sub.1, Qt.sub.2, Qt.sub.3, Qt.sub.4 Transfer field effect transistor (pass gate transistor) [0229] 10 Semiconductor layer [0230] 11 Isolation region [0231] 12a n-type well region [0232] 12b p-type well region [0233] 15 Gate insulating film [0234] 16d, 16p, 16t Gate electrode [0235] 17d.sub.1, 17p.sub.1, 17t.sub.1 First main electrode region [0236] 17d.sub.2, 17p.sub.2, 17t.sub.2 Second main electrode region [0237] 21 Interlayer insulating film [0238] 22a, 22b, 22c, 22d, 22e Conductive plug [0239] 23a First electrode [0240] 23b Insulating film [0241] 23c Second electrode [0242] 24 Interlayer insulating film [0243] 25a.sub.1, 25a.sub.2, 25c.sub.1, 25c.sub.2, 25d.sub.1, 25d.sub.2, 25e.sub.1, 25e.sub.2 Relay wiring [0244] 26 Interlayer insulating film [0245] 27 Conductive plug [0246] 28c.sub.1, 28c.sub.2 Ground wiring [0247] 28d Power supply line [0248] AL Axon line [0249] BL.sub.1 Bit line [0250] BL.sub.2 Bit line bar [0251] DL.sub.1 Dendrite line [0252] DL.sub.2 Dendrite line bar [0253] WL Word line [0254] Qd.sub.1, Qd.sub.2, Qd.sub.3, Qd.sub.4 Drive field effect transistor (drive FET) [0255] Qp.sub.1, Qp.sub.2 Load field effect transistor (drive FET) [0256] Qt.sub.1, Qt.sub.2, Qt.sub.3, Qt.sub.4 Transfer field effect transistor (transfer FET) [0257] Qd.sub.5, Qd.sub.6 Drive tunnel field effect transistor (drive tunnel FET) [0258] Qt.sub.5, Qt.sub.6 Transfer tunnel field effect transistor (transfer tunnel FET) [0259] PD.sub.1, PD.sub.2 Parasitic diode