DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT OF THE INVENTION

[0023] Referring now to the drawings, and more particularly to FIGS. 1A and 1B , there is shown an example of the prior art of a planar static random access memory (SRAM) arrays. FIG. 1A is a schematic circuit diagram illustrating a standard six transistor CMOS SRAM cell. The NMOS transistors Q 1 and Q 2 are the “access” devices, the NMOS transistors Q 3 and Q 4 are “driver” transistors, and the two PMOS transistors Q 5 and Q 6 are “load” transistors.

[0024] The layout plan view of a standard SRAM cell is shown in FIG. 1B . Q 1 and Q 2 NMOS transistors are formed by the overlap of the active n-silicon 1 and polysilicon layers 3 and 4 . The source contacts 10 of Q 1 and Q 2 are contacted by metal layer 6 to V SS or ground. Similarly, Q 3 and Q 4 are formed by overlap of the active silicon 31 and polysilicon layer 5 which forms the word line (WL). The drain contacts of Q 3 and Q 4 are connected to bitlines {overscore (Bit)} and Bit, respectively. The PMOS Q 5 and Q 6 transistors are connected to metal layer 7 to V DD . The drain contacts 40 and 42 of Q 5 and Q 6 are connected to the nodes, shown as solid dots in FIG. 1A .

[0025] The prior art in SRAM cells fabricated in three-dimensions (3D) using thin film Si PMOS load transistors is shown in FIGS. 2A and 2B . The circuit of FIG. 2A is substantially the same as that of FIG. 1A .

[0026] FIG. 2B shows the structure of the prior art in three-dimensional (3D) SRAM cells, a fabrication method used to increase the density of memory arrays. The three-dimensional SRAM cell places the PMOS transistors, Q 5 and Q 6 , in a thin film transistor (TFT) layer, preferably polycrystalline Si (p-Si) made by the excimer laser annealing method. Transistors Q 1 to Q 4 are made in the crystal Si wafer substrate. More particularly as shown in FIG. 2 B, Q 1 and Q 2 NMOS transistors are formed by the overlap of the active n-silicon 21 and 23 and first polysilicon layers 16 and 15 , respectively. Similarly, Q 3 and Q 4 are formed by the overlap of the active silicon 11 and 12 and first polysilicon layer 17 which forms the word line (WL). The drain contacts 18 of Q 3 and Q 4 are connected to the bit lines ({overscore (BL)} and BL) which are formed in the aluminum (Al) metal layer. The second polysilicon layer 28 forms the gate of PMOS TFTs Q 5 and Q 6 (bottom gate TFT structure). The third polysilicon layers 13 and 14 forms the active layer of TFTs Q 5 and Q 6 and also forms the V DD line. The overlap of the second polysilicon layer 28 and third polysilicon layers 13 and 14 form the TFTs Q 5 and Q 6 , respectively.

[0027] Differential cascode voltage switch (DCVS) logic is a double-rail CMOS circuit technique which has potential advantages over conventional single-rail NAND/NOR random logic in terms of layout area, circuit delay, power dissipation and logic flexibility. DCVS is constructed of stacked NMOS differential pairs which are connected to a pair of cross-coupled PMOS loads for pull up. No direct current (d.c.) is drawn in static mode. Therefore, complicated Boolean logic functions which may require several gates in conventional CMOS logic can be implemented in a single stage gate in DCVS.

[0028] The prior art in a simple differential cascode voltage switch (DCVS) logic relevant to the present invention is shown in FIGS. 3A and 3B . The schematic circuit diagram of conventional DCVS AND/NAND gate is FIG. 3 A, and the layout plan view is shown in FIG. 3B . In this case, all six transistors are made in a single level of the Si wafer substrate. There are four driver transistors, Q 1 to Q 4 , which are NMOS devices and these form the n-channel logic evaluation (true and complement) trees. The circuit load is formed by two cross-coupled PMOS load transistors, Q 5 and Q 6 , which are significant to the present invention because these devices occupy a large area of the Si wafer substrate and thus prevent the conventional DCVS logic from achieving very high area density.

[0029] In FIG. 3 A, the left leg of NMOS logic tree is constructed of two NMOS transistors Q 2 and Q 1 in series and a connection to ground to form one pull-down network. Q 2 and Q 1 act as a switch which controlled by their gate signals A and B, respectively. The right leg of NMOS logic tree is constructed of two NNOS transistors Q 3 and Q 4 in parallel. Both transistors are connected to ground to form another pull-down network. Q 3 and Q 4 are controlled by their gate signals, complementary inputs {overscore (A)} and {overscore (B)}, respectively. The pull-up network is constructed of two cross-coupled PMOS transistors, Q 5 and Q 6 . When the input signals A and B swing from low to high, transistors Q 1 and Q 2 turn ON. The node {overscore (Y)} is then discharged to ground. The node Y is floating at the transition period while the complementary input signals {overscore (A)} and {overscore (B)} swing from high to low.

[0030] Both of the NMOS transistors Q 3 and Q 4 are OFF. The ground level on the node {overscore (Y)} turns the cross-coupled PMOS load transistor Q 6 ON. The output node Y will be charged high. This realizes the dual AND/NAND logic functions.

[0031] The layout plan view of a simple DCVS AND/NAND gate is shown in FIG. 3B . Q 1 and Q 2 NMOS transistors are formed by the overlap of the active n-silicon 31 and polysilicon layers, 36 and 37 . The source and drain diffusion is thereby self-aligned to the gates A and B. The source contact of Q 1 is connected by metal layer 41 to V SS or ground. The drain contact of Q 2 , 43 , is connected to node {circle over ( 1 )} or {overscore (Y)}. Similarly, Q 3 and Q 4 NMOS transistors are formed by the overlap of the active silicon 31 and polysilicon layers, 38 and 39 . The source and drain diffusion is thereby self-aligned to the gates {overscore (A)} and {overscore (B)}. The source contacts of Q 1 , 45 and 47 , are connected by metal layer 41 to V SS or ground. The shared drain contact of Q 3 and Q 4 , 49 , is connected to node {circle over ( 2 )} or Y. Q 5 and Q 6 PMOS transistors are made in a p+ region implanted into n-well region 33 . The n-well is usually a deeper implant compared with the transistor source/drain implants; therefore, it is necessary for the outside dimension to provide sufficient clearance between the n-well edges and adjacent n+ diffusion. Again, Q 5 and Q 6 PMOS transistors are formed by the overlap of the active p- silicon 32 and polysilicon layers, 34 and 35 . The source contacts of Q 5 and Q 6 , 51 and 53 , are connected to metal layer 50 to V DD . The drain contact of Q 5 and Q 6 , 55 and 57 , are connected to node {circle over ( 1 )} and {circle over ( 2 )}, or Y and {overscore (Y)}, respectively.

[0032] A cross-sectional view of the structure illustrating the present invention is shown in FIG. 4 . More particularly, FIG. 4 shows a schematic cross-section of the most general form of the present invention, a three-dimensional (3D) CMOS transistor pair used to form both logic and SRAM memory elements. For the sake of clarity, only the transistor levels are shown (wiring levels are not shown). In this simplified cross-section, a single NMOS transistor, 400 , is made in the crystalline Si wafer substrate, 401 . A PMOS load transistor, 411 , is made in a Si layer above the NMOS device. The Si layer, 406 , used for the PMOS TFT is preferably polycrystalline Si (p-Si) made by the excimer laser annealing method. Alternatively, this is p-Si made by the rapid thermal annealing (RTA) method. Briefly, lower parts of the structure are a thick insulator, 402 , via holes, 403 , and conductors, 404 , filling the via holes.

[0033] The thick insulator, 402 , is planarized by chemical mechanical polishing (CMP) methods to leave a flat surface, 405 , for subsequent fabrication of the PMOS transistor 411 . Upper parts of the structure are a thin film Si layer, 406 , a gate dielectric layer, 407 , a gate conductor, 408 , and source and drain contacts 409 . The source and drain metal level, 409 , is insulated by the thick insulator (passivation) layer, 410 . Fabrication of the thin film Si upper level of this structure is described in detail below in reference to FIGS. 8A to 8 D.

[0034] A general case and a preferred embodiment of the present invention is now described in reference to FIGS. 5A and 5B which show a detailed structure of one embodiment of the present invention, a 3D circuit construction of AND and NAND differential logic gates in DCVS logic. The DCVS circuit concept in its differential form is illustrated in FIG. 5 A, which shows a circuit schematic diagram for both AND and NAND gates in DCVS logic (three-dimensional construction, or 3D DCVS). Again, for sake of clarity, only the transistor levels and selected wiring levels up to M 4 are shown (complete wiring levels are not shown). Active transistors Q 1 to Q 4 are made in the crystal Si wafer substrate. The two cross-coupled PMOS road transistors Q 5 and Q 6 are made in a TFT layer, preferrably using polycrystalline Si (p-Si) made by the excimer laser annealing method. Depending on the differential inputs, one output (either F or {overscore (F)}) is pulled down by the NMOS combinatorial logic evaluation tree network. Positive feedback action sets the PMOS latch to static output F and {overscore (F)} or full differential V DD and ground logic levels.

[0035] The basic circuit operation of 3D DCVS is the same as 2D DCVS which was described above with reference to FIG. 3A . In the 3D case, the pull-up load network consists of two cross-coupled PMOS TFTs. This provides significant advantages in design flexibility for said load devices. The pull-up performance, i.e., a faster rise time, of a complex logic gate can be dramatically enhanced. It should be noted that the double-rail logic has been used exclusively in advanced high performance digital systems. The design procedure to construct a more complex NMOS logic tree for 2D or 3D DCVS can be synthesized by the Karnaugh map (K-map).

[0036] The detailed cross section showing fabrication of this circuit is in FIG. 5B . The p- epitaxial layer 501 is deposited on the P+ substrate 500 . A standard NMOS process is used to make the active transistors Q 1 to Q 4 on the P+ substrate, 500 . The active area 503 for transistors Q 1 to Q 4 is defined by ion implant of the N-dopant. Then the shallow trench isolation (STI), 502 , isolates adjacent devices Q 2 and Q 3 . A deposited polysilicon layer is patterned to form self-aligned Si gates, 524 , 525 , 526 and 527 , of transistors Q 1 to Q 4 , respectively. Ion implantation is used to form the N-doped source and drain regions, 503 . The source and drain contacts 505 are formed and connected to the first metal layer (M 1 ). The source junction contacts of transistors Q 1 , Q 3 and Q 4 are connected to M 1 , ground. The transistor gates of Q 1 , Q 2 , Q 3 , and Q 4 are connected to input signals A, B and {overscore (A)}, {overscore (B)}, respectively. A thick insulator, 506 , is deposited by chemical vapor deposition (CVD). As mentioned previously, the thick insulator, 506 , is planarized by chemical mechanical polishing (CMP) methods to leave a flat surface, 518 , for subsequent fabrication of the PMOS load transistors.

[0037] Important via holes for connections between the bulk NMOS transistors and PMOS thin film transistors (TFTs) are then patterned and etched. These via holes are filled with the conductors, 530 and 532 . Conductor 530 connects Q 2 to Q 5 . Conductor 532 connects Q 3 and Q 4 to Q 6 .

[0038] PMOS load transistors are made in a TFT Si layer, preferably polycrystalline Si (p-Si) made by the excimer laser annealing method. The structure begins with deposition of the thin film Si layer, and patterning into active islands 507 . The conformal deposited gate insulator layer 508 is made. Next, deposit highly doped polysilicon layer to form self-aligned silicon gate, 509 . Ion implantation is used to form the P-doped source and drain regions. The source and drain contacts are connected to the M 2 or M 3 metal layers. The drain contact 531 of TFT Q 5 is connected to metal layer M 2 , forming node {overscore (F)}. The drain contact 533 of TFT Q 6 is connected to metal layer M 2 , forming node F. These nodes are connected to output signal lines {overscore (F)} and F, respectively. Furthermore, the source contact 512 of TFT Q 5 and contact 511 of TFT Q 6 are connected to M 3 layer, 514 and then through via 516 to the fourth metal layer (M 4 ), 517 . V DD is applied from thin film wire 514 through interconnect 511 to TFT Q 6 . V DD is applied from thin film wire 514 through interconnect 512 to TFT Q 5 . Deposited dielectric layers 510 and 515 isolate thin film wire levels. Only essential wire levels are shown. only one of the wiring levels located above V DD 514 is shown. Standard VLSI technology is used to form the remaining back end of the line connections.

[0039] A specific case and a preferred embodiment of the present invention is now described in reference to FIGS. 6A and 6B . A new high-performance and low power circuit technique called 3D-differential cascode voltage switch with pass-gate (DCVSPG) logic family is described. The circuit style is designed using pass-gate logic tree in DCVSPG instead of the NMOS logic stacked tree in DCVS. The DVCS is classified as a ratio circuit. The DCVSPG is considered as a ratioless circuit. FIG. 6A is the circuit diagram for a simple AND/NAND gate made using DCVSPG logic in a three-dimensional structure.

[0040] In FIG. 6 A, the left-side leg of pass-gate the logic tree is constructed of two NMOS Q 2 and Q 1 in parallel. Note that in DCVS these two NMOS transistors are in series ( FIG. 5A ), but in DCVSPG these two NMOS transistors are in series. The advantages will be obvious when complex logic functions are designed (see FIGS. 7A to 7 D). The right-hand leg of the pass-gate tree is constructed of two NMOS Q 3 and Q 4 in parallel as well. The pass-gate logic tree can be synthesized in a very systematic way by recursively using a Karnaugh map. The basic logic with two input variables A and B is shown in FIG. 6A . The input signal A or B can be either the NMOS gate control or NMOS source connection. In this case, if we assume the signal A is the control variable, the B signal will be the function variable. The control variable is used to connect to the gate and the function variable is connected to the source of the NMOS device. Under the control signals {overscore (A)} and A, we grouped the terms together as shown in FIG. 6A . {overscore (A)} connects to the control gates of both Q 1 and Q 3 . A connects to the control gates of both Q 2 and Q 4 . The source of Q 1 , Q 2 , Q 3 and Q 4 are connected to the functional variables V DD {overscore (B)}, ground and B, respectively. Two cross-coupled TFT PMOS transistors, Q 5 and Q 6 , make up the pull-up network, exactly as in Figure SA.

[0041] The 3D DCVSPG AND/NAND circuit shown in FIG. 6A actually solves the floating node problem by replacing the NMOS tree with the pass-gate design. with the same previous state, when both input signals A and B swing from low to high, Q 2 and Q 4 both turn ON. The node {overscore (F)} is then discharged into ground when the complementary signals {overscore (A)} and {overscore (B)} swing from high to low. However, the output node F is charging up to high sate immediately. This realizes the dual AND/NAND logic functions. No floating node problems occur.

[0042] FIG. 6B is the detailed cross-section of the structure implementing the circuit of FIG. 6A . For sake of clarity, only the transistor levels and selected wiring levels up to M 4 are shown (complete wiring levels are not shown). The detailed fabrication of this circuit is very similar to FIG. 5B . The only modification is the transistor connection scheme. In FIG. 6 A, the left-side leg of NMOS network is now a parallel connection. The p-epitaxial layer 601 is deposited on the P+ substrate 600 . A standard NMOS process is used to make the active transistors Q 1 to Q 4 on the P+ substrate, 600 . The formation of this structure was described above with reference to FIG. 5B . The shallow trench isolation (STI), 602 , isolates adjacent devices Q 2 and Q 3 . A deposited polysilicon layer is patterned to form the self-aligned silicon gates, 604 , 624 , 625 and 626 . Ion implantation is used to form the N-doped source and drain regions, 603 .

[0043] The four following connections are made using the first metal layer (M 1 ). The source junction contacts of transistors Q 1 and Q 2 are applied to V DD and {overscore (B)}, respectively. The source junction contacts of transistors Q 3 and Q 4 are applied to ground and B, respectively. These connections are made using the deposited polysilicon layer. The transistor gate of Q 1 , 604 , and the transistor gate of Q 3 , 625 , are connected to input signal {overscore (A)}. The transistor gate of Q 2 is connected to input signal A, and the gate of Q 4 is also connected to input signal A. Then, a thick insulator, 606 , is deposited by CVD. As mentioned previously, the thick insulator, 606 , is planaraized by chemical mechanical polishing (CMP) methods to leave a flat surface, 618 , for subsequent fabrication of the PMOS load transistors. Via holes are patterned and etched, and the conductors, 630 , are deposited filling via holes.

[0044] Again, the formation of the TFTs Q 5 and Q 6 can be followed beginning with the active area for transistor islands 607 . Then deposit the thin insulator gate dielectric layer 608 . Next, deposit a highly doped polysilicon layer to form self-aligned silicon gate, 609 . Ion implantation is used to form the P-doped source and drain regions using the gate 609 as a self-aligned mask. The drain contacts of TFTs Q 5 and Q 6 , 63 and 632 , are connected to output signal line {overscore (F)} and F, respectively, using the second metal layer (M 2 ). Furthermore, the source contacts of TFTs Q 5 and Q 6 , 612 and 611 , are connected to the third metal layer (M 3 ) layer, 614 . A connection of M 3 to the fourth metal layer (M 4 ), 617 , is shown as stud 616 . V DD is applied from thin film wire 614 through interconnect 611 to TFT Q 6 . V DD is applied from thin film wire 617 through interconnect 616 to TFT Q 6 . Deposited dielectric layers 610 and 615 isolate the thin film wire levels. Only essential wire levels are shown. Only one of the wiring levels located above V DD 614 is shown. Standard VLSI technology is used to form the remaining back end of the line connections.

[0045] The most general form of the present invention is now described in reference to FIG. 7A which shows a general form of the present invention comprising a logic gate with multiple differential (dual rail) inputs, forming a combination logic network. Two cross-coupled PMOS TFTs 76 and 77 are at the top. Logic design leverage is achieved by DCVS or DCVSPG by cascading differential pairs of NMOS devices into the powerful combinatorial logic tree networks capable of processing complex Boolean logic functions. Therefore, complicated logic which may require several gates in conventional CMOS can be implemented in a single stage gate in DCVS or DCVSPG. For example, as shown in FIGS. 7B, a logic sum circuit can be implemented with sixteen transistors (eight PMOS and eight NMOS transistors) in conventional CMOS circuit. On the other hand, in DCVS twelve transistors (two PMOS and ten NMOS transistors), as shown in FIG. 7 C, and in DCVSPG, ten transistors (two PMOS and eight NMOS transistors) make the logic gate circuit, as shown in FIG. 7D .

[0046] Combinatorial logic devices can be designed with unstacked PMOS devices used sparingly as pull-up devices in load and buffer circuitry. Optimization of the PMOS devices and optimization of the PMOS to NMOS spacing can therefore be relaxed, relieving the device and process complexity burden for DCVS design.

[0047] The single set of process steps to make both logic and memory structures according to the present invention is now described in detail in reference to FIGS. 8A to 8 D which show the general flow of process steps to make the Thin Film Transistor (TFT) PMOS levels of the present invention. We begin by assuming a level of completed NMOS transistors, 802 , is existing on the Si wafer substrate, 801 . A thick insulator, 803 , is deposited and planarized by chemical mechanical polishing (CMP). This provides a smooth starting surface 800 for TFT fabrication. Via holes, 804 , are patterned lithographically, etched, and filled with a conductor to form the interlevel connect, 805 . This is preferably a refractory metal, for example tungsten. After planarization of the connect, 805 , a layer of amorphous Si (about 500-1,000 Angstroms (Å) thick) is deposited by a suitable method (sputtering, plasma enhanced CVD, or LPCVD), patterned lithographically into islands, and converted to p-Si. Excimer laser annealing (ELA) is the preferred method to crystallize the TF Si, although rapid thermal annealing (RTA) may be used. FIG. 8A shows the resulting poly-Si island, 806 .

[0048] FIG. 8B shows deposition of the gate dielectric, 807 , which is preferably amorphous SiO 2 about 1,000 to 1,500 Å thick, and is deposited by chemical vapor deposition (CVD) at a temperature of 300 to 400° C. The gate electrode is deposited as a blanket metal layer (aluminum, or other metal), and patterned lithographically to form the gate, 808 . The P-type dopant boron, 809 , is placed in the TF Si layer, 806 , using ion implantation or ion shower doping. The energy of the B+ ions is selected so the ions penetrate through the dielectric, 807 , and into the TF Si layer, 806 . The gate, 808 , may be used to mask the layer where no dopant is placed, so the gate may be a self aligned mask. Optionally, a two step doping procedure may be used to form a lightly doped drain structure. Then, the structure is heated for a few seconds by the RTA or the ELA methods to activate the dopant boron atoms.

[0049] FIG. 8C shows deposition of a thick insulator, 810 , which is amorphous SiO 2 or silicon nitride deposited by chemical vapor deposition. Via boles, 811 , are patterned and etched to contact the source and drain regions of the TFT, 812 .

[0050] FIG. 8D shows deposition of conductors, 813 , filling the via holes, 811 . The preferred material is a aluminum, although other metals can be used. A source/drain metal level, 814 , is deposited and pattered into thin film wires (the TFT source/drain metal level). Finally, a passivating insulator, 815 , is deposited. The wiring levels, 814 , are not shown in detail because we emphasize here the circuit diagrams and the TFT layer. The essential wiring levels are shown in FIGS. 5B and 6B .

[0051] While the invention has been described in terms of preferred embodiments, those skilled in the art will recognize that the invention can be practiced with modification within the spirit and scope of the appended claims.