Reconfigurable Memory Circuit

1. Field of the Invention

The field of the present invention relates to an improved method and apparatus for interconnection of microcircuit elements on a wafer to enable high reliability wafer scale integration and a new improved memory circuit.

2. Prior Art

Present methods of semiconductor fabrication typically require that a plurality of identical semiconductor elements or circuits be fabricated on a single wafer substrate through a series of process steps. At the conclusion of the fabrication process, the elements or circuits are tested and the defective elements identified. The wafer is then scored and diced into individual parts, each containing a complete circuit. These operable parts are then packaged to provide external connections and appropriate environmental protection for the finished product.

It has long been realized that circuit costs as well as space and power requirements could be reduced if the operable parts or elements on the wafer could be interconnected on the wafer itself. This would reduce the necessity for circuit boards or interconnection wiring between elements. In addition, dicing would be eliminated and packaging would be relatively simple for the single wafer itself.

The variety of methods for interconnecting elements on a wafer which have been suggested these include a metalization step after the circuits or elements of the wafer are tested to interconnect the operable elements on the wafer. The disadvantage of this method is that a different mask is required for each individual wafer, due to the fact that not only the yield may be different for a given wafer but defects are randomly created. The production of a unique mask for each wafer not only greatly increases the cost of the finished product but is considerably time consuming thereby slowing the manufacturing process. Other wafer integration schemes such as described in U.S. Pat. No. 3,940,740, provide for spare rows and columns on the wafer matrix for appropriate sparing of defective elements. The disadvantage of this configuration is that an entire row of elements must be provided to spare a single defective matrix element. Also patents such as U.S. Pat. No. 4,092,733 provide schemes where multiple elemental structures may be interconnected. However, no particular circuitry is described to implement any particular memory structure. A preferred method would be one where spare memory elements may be inserted in the matrix on a random basis to allow all spare elements to be used, a technique which has not yet been perfected in the prior art.

This invention utilizes a reconfigurable memory circuit having an associated interconnect circuit with each memory cell within a memory array. Appropriate signals to the memory cell interconnect circuit will either engage the memory cell into communication with a common bus line or disengage defective memory cell out of such communication by effectively switching in a bypass path for data around the defective memory cell. As described before the prior art systems tend to be bulky, the actual area of the microcircuits themselves being far less than that of the printed circuit boards upon which they are mounted. They are also relatively expensive due to the large number of components and manufacturing steps both manual and automatic involved therein. Likewise, the number of bonds, both internal to the microcircuit packages and external from package to package connections result in decreased reliability. The efforts that have been made to interconnect the individual, that is, the chip level large scale integrated (LSI) microcircuits on their common wafer substrate have previously been proven costly and impractical. In addition, they share a common fault in that the interconnects cannot be repaired physically or can be repaired only at great cost. In military systems which require an improved reliability level, fault tolerant or redundant techniques are typically utilized. Such techniques indeed produce reliable military systems, but they entail significant cost and size penalties. An example, of course, is the triply redundant majority logic implementation of high reliability systems where a more than threefold increase in component hardware is required. This evidences itself as a two for one parallel arrangement. Should the same corresponding cell location fail twice the cell becomes inoperative. However, the instant invention corrects the failure of any number of cells and allows the switching of bad memory cells out of the array and a switching of good memory cells into the array. Upon the utilization of all spares available a reduction of the length of the memory array itself is possible while maintaining good continuous or sequential memory cells. Wafer scale integration (WSI), is a method for greatly decreasing the number of leads and bonds necessary to interconnect a given package with external electronics, however, unless there is an adaptive technique available WSI yield problems are insurmountable and require the inclusion of redundant circuits on the wafer thereby decreasing the yield per unit area of actual circuitry used. In the adaptive integration (AI) approach of the reconfigurable memory circuit of the instant invention, the good circuits are located by testing and are then connected to the interconnect bus by electrically alterable interconnect circuits to produce a complete subsystem on a single wafer ready for packaging. Not only does this AI approach eliminate the wafer die and break, die sorting, multidie and wire bonding steps, they also increase the yield of the microcircuits themselves per unit area on the wafer while providing improved reliability due to the reduction of the number of mechanical interconnection as well as adaptive sparing. A minimum number of spare microcircuits is set aside on a wafer. On detection of failure the failed component is by passed and the spare serially connected to the system bus. Thus sparing redundancy and high system reliability is accomplished with a minimum of hardware cost and little increase in dedicated circuitry. In the instant invention a memory cell which comprises a series of subcells comprising at least one static two bit register with two bits of nonvolatile MNOS memory in parallel with the two bits of the shift register in series with a plurality of such cells each having its own interconnect circuit associated with the individual memory cell. Data is shifted through the shift register forward one bit for each shift pulse. When data is positioned within the memory cell at selected MNOS storage sites the MNOS transistors are cleared by an erase pulse and a subsequent write pulse will copy data from the shift register into nonvolatile storage. A load pulse will copy data from nonvolatile memory into the shift register. The shift, load, erase, command effect all memory cells.

The further advantages of this invention may be readily noted with reference to the drawings.

FIG. 1 is a schematic block diagram of a plurality of interconnected memory cells.

FIG. 2 is a logic diagram of the interconnect switch associated with each memory cell.

FIG. 3 is a circuit diagram of the interconnect switch and associated logic.

FIG. 4 is a logic diagram of a memory cell having a plurality of submemory cells in parallel with a shift register.

FIG. 5 is a circuit diagram of the submemory cell, shift register and associated logic.

FIG. 6 is a circuit diagram of the output buffer of the memory cell.

FIG. 7 is a graphical representation of the timing required for cell operation.

With reference to the drawings FIG. 1 illustrates a plurality of memory cells 10, although the particular cells could be a series of any particular pipe-lined serial circuitry, connected in series through an interconnect switch 12. Only three such configurations are shown although it is understood that any number of such connections may be made on a wafer limited only by the size of the particular substrate. In this particular configuration the center illustrated interconnect switch is set to the bypass state, that is, data in on line 20 will bypass the associated memory cell 10 on line 22 by virtue of the fact that the interconnect switch senses data only on the bypass line 22 and ignores data output from the memory cell on output line 24. An examination of FIG. 2 shows a detailed logic diagram of the memory cell 10 and the interconnect switch 12 configuration. Interconnect switch 12, defined by the dotted line, contains a switch flip-flop with a nonvolatile MNOS memory cell. The nonvolatile MNOS memory cell insures the configuration stability of the interconnect switch. Therefore, with the MNOS memory cell being independent of power transients and power failures, once the interconnect switch is set, it remains in that pass, high (B), or bypass, high (B), state. The pass or bypass output states are complementary, that is, when one is in an electrically low state the other is in an electrically high state. Each of these outputs is input to an individual "NAND" gate. B going to "NAND" gate 30 and B going to "NAND" gate 32. Output from the memory cell on line 24 is connected to "NAND" gate 32. The bypass line 22 is connected to "NAND" gate 30 the output of "NAND" gate 30 and 32 is then connected to "NAND" gate 34 which outputs on output line 36. As may be noted, when B is equal to 0, B will be equal to 1, this forces the output of gate 32 to correspond to the inverse of the output of the memory cell on line 24. Gate 30, with B input at 0, on the other hand, will not pass whatever information appears on the bypass line 22 to the "NAND" gate 34. "NAND" gate 34 will again invert that input from line 24 and output data on line 36 which is representative of the data passing through the memory cell. A reverse of the state of the switch flip-flop causing B to be equal to 1 and B equal to 0 will ignore the input on line 24 and pass data on line 22. Switch flip-flop 50 has three control lines, a B input line (BC) a B input line (BC) and an erase input line. Switch flip-flop 40 contains MNOS type memory cells and before a given state can be set into the switch flip-flop, an ERASE pulse (Vesw) must appear on erase line 42 to clear the MNOS memory, thereafter either the B or the B input lines may be momentarily grounded to set the flip-flop into the particular state desired. That is, if output B is to be high thereby bypassing the memory cell a ground should be placed on the B input line BC. This will set the output of B equal to 1 and the output of B equal to 0 effecting the bypass of the memory cell as described before. If the reverse is desired then the B input line BC must be grounded thereby setting B output to a 1 and B output to a 0 thereby accepting information from the memory cell.

More specifically as shown in FIG. 3, the MNOS transistors Q3 and Q4 must be erased, that is, placed in their low conductivity state prior to writing the desired information into the MNOS transistor. Accordingly, the V.sub.DD /V.sub.MM interconnect switch power bus is grounded while an ERASE pulse is applied to the erase line 42 which is connected to the substrate. After the prescribed erase pulse (Vesw) is applied to the ERASE line Q3 or Q4 may be written by grounding a selected Bypass BC or Pass BC line while leaving the unselected line at open circuit, and re-establishing the voltage at line V.sub.DD /V.sub.MM to V.sub.DD. The normal condition for the Bypass BC and Pass BC lines is open circuit. The Bypass Control (BC) is line selected (grounded) to bypass the associated memory cell or the Pass Control (BC) is selected (grounded) to include the associated memory cell on the data path. After selecting the appropriate Bypass or Pass line, a minus 30 volts (V.sub.MM) SWITCH WRITE pulse is applied to the V.sub.DD /V.sub.MM power bus which places the selected MNOS transistor in a low conductance state. This completes the ERASE/WRITE cycle for the interconnect switch cells. All grounded Bypass and Pass inputs are then returned to their normal open circuit condition. Minus 15 volts (V.sub.DD) is applied to the V.sub.DD /V.sub.MM bus and normal chip operation is resumed.

FIG. 3 shows the circuit configuration for the interconnect switch and the associated logic. Q3 and Q4 are MNOS transistors which provide the nonvolatile memory for the interconnect switch. Q5 and Q6 are enhancement mode transistors which latch the state of the switch flip-flop and output the bypass BC and Pass BC states on the B output line and B output line, respectively connected to enhancement mode transistors Q9 and Q12. Transistors Q1, Q2, Q17 and Q18 are bias load devices and Q7 provides current limiting functions for the MNOS transistor gates. Q19 acts as voltage protect device.

Q1, Q2, Q7, Q8, Q11, Q14, Q17, Q18 and Q19 are all depletion mode transistors used primarily to control current in the particular circuit configuration. Q8, Q9, Q10 and Q11, Q12, Q13 and Q14, Q15, Q16 form the three "NAND" gates 30, 32 and 34 required for the interconnect previously described.

FIG. 4 is a logic diagram of a particular memory cell. Only two MNOS nonvolatile memory subcircuits 100 defined by a dotted line are shown along with the two bit parallel shift register for clarity, although numerous such subcells could be interconnected. An erase line 60 load line 62 and a write line 66 are shown interconnected with the subcircuit 100. The data latch of the memory subcircuit, logically represented by the interconnected NAND gates 72 and 74, are also functionally shared as the shift register latch. Proper timing signals are illustrated in FIG. 7.

FIG. 5 is a detailed circuit diagram of the MNOS memory subcircuit. Transistors Q27 and Q35 are the MNOS load transistors which, in connection with the transistors Q23 and Q24 constitute the nonvolatile memory cell. Q29 and Q37 couple input from the MNOS transistors to the latch on receipt of the LOAD (V.sub.L) pulse through enhancement mode transistors Q28 and Q26. Enhancement mode transistors Q46 and Q47 couple the output of Q23 and Q24 to the inputs of Q23 and Q24 thus forming the latch portion of the circuit. Enhancement mode transistors Q40 and Q41 couple the input from the prior stage into the latch circuit when it is functionally part of the shift register circuit. Depletion mode transistors Q41, Q42, Q30, Q34, Q38, Q42, Q43, Q44 and Q45 provide current limiting functions within the circuit. Q26, Q28 and Q36 couple the LOAD pulse (V.sub.L) with the latch circuit. Q26 also turns on Q25 and Q33 which are enhancement mode transistors thereby driving current through the MNOS transistors Q27 and Q35 respectively. Depletion mode transistor Q31 is voltage protect device. Enhancement mode transistors Q48 and Q49 provide coupling of the information in the latch Q3 and Q4 through Q27 and Q26 into the MNOS transistors on receipt of a WRITE pulse (V.sub.W). ERASE (V.sub.E) operates as in the interconnect switch.

FIG. 6 is a circuit diagram of output buffer driver 26 and is simply a two-stage enhancement mode transistor amplifier where in Q62 and Q64 form the two-stage amp. Q61 and Q63 provide current limiting functions in the circuit. Input line 36 and output line 28 are shown on the diagram for clarity.

FIG. 7 illustrates the timing signals to be applied to the disclosed apparatus for all modes of operation herein described.