Entire Wafer Stress Test Method for Integrated Memory Devices and Circuit Therefor

1. Field of the Invention

The present invention relates to the testing of memory devices and more particularly to the testing of static random access memory (SRAM) devices, first-in first-out (FIFO) memory devices, and other memories incorporated in integrated circuits.

2. Description of the Prior Art

The premature or infant failure of memory cells in integrated SRAMs, multiple port memories, FIFO memories and other memory products has been an unfortunate but all too common occurrence. Moreover, failure of such parts in the field is unacceptable to vendors of products incorporating integrated circuit memory. As a result, nondestructive testing of all integrated circuits to expose and detect that integrated circuits subject to infant failure is dictated by the market.

One contemporary testing regimen is to place memory devices into burn in oven, elevating the device temperature, and then exercising this devices by varying applied voltages to the product. For some large capacity memory devices the period in the burn in oven has reached 96 hours, in order to stress each of over one thousand wordlines and millions of memory cells. Such long burn-in cycles pose an obvious bottleneck to production, and are useless for generating up to the minute information about possible faults in the manufacturing process. An accelerated stress mode that eliminates this bottleneck without damaging good memory product would have apparent benefits.

For some test or operating modes of the memory array, selection of all or a portion of row and/or bit lines at a single time is desirable. An example of a test mode where selecting all or a portion of the row and bit lines at a single time is used is described in co-pending U.S. patent application Ser. No. 07/954,276, entitled Stress Test For Memory Arrays In Integrated Circuits, filed Sep. 30, 1992, assigned to SGS THOMSON Microelectronics, Inc. and incorporated herein by this reference. A plurality of rows is selected at one time and a stress voltage is placed on a plurality of bit and complementary bit lines. In this manner the memory cells within the memory array are stress tested in order to detect latent defects.

Another example of test mode, where selecting all the row and bit lines at a single time is used, is described in U.S. Pat. No. 5,341,336 entitled Method For Stress Testing Decoders And Periphery Circuits, assigned to SGS THOMSON Microelectronics, Inc. and incorporated herein by this reference. A plurality of rows and bit lines are selected or deselected simultaneously and a stress voltage is applied to the integrated circuit. In this manner latent defects within decoders and periphery circuits can be detected.

A circuit that allows for the simultaneous selection or deselection of a plurality of rows and columns within a memory array is described in U.S. Pat. No. 5,339,277 entitled Address Buffer, assigned to SGS THOMSON Microelectronics, Inc. and incorporated herein by this reference. A first and a second circuit generate a true and a complementary signal, respectively, during normal operations of the integrated circuit. When desired the first and second circuits may be used to generate two signals of the same voltage level. The two signals of the same voltage level may then be used by an address decoder to simultaneously select or deselect a plurality of rows and/or columns within a memory array.

The increasing complexity of the memory devices is also increasing the number of signal lines that must be controlled during the test and consequently the complexity and the cost of test equipment.

Therefore, it would be desirable to provide a circuit and a method for stress testing integrated memory circuits at wafer level that permits to minimize the number of signals used to control the integrated circuit during the test and that allows parallel testing of a plurality of integrated circuit dies on a single wafer. Such a circuit and method would permit parallel testing of all dies on a wafer (e.g. 600 dies/wafer) at the same time with a 600.times. reduction in test time and also avoid consequent burn in oven insertion of packaged devices, resulting in an important decrease of the test time and of the complexity of test probes and test equipment and verifying reliability of bare silicon dies.

A circuit and related method are provided for parallel stressing of a plurality of memory circuits integrated on dies on a silicon wafer. A test mode control circuit, having a first and a second test mode control input, is used, during special test operation mode, to force outputs of address buffers, data buffers and other signal buffers, like chip-enable or write buffers, to predetermined logic values so that all row and column decoders are selected and predetermined data is written into the array of memory cells. Contemporaneously are also exercised entire paths of buffers. The integrated circuit is then heated and maintained at an elevated temperature for a desired time, and then cooled down. In this way it is possible to stress test for ionic contamination, trap sites and weak oxides one or more integrated circuits on the same wafer in a short time, requiring only a limited number of test signals. It is possible, indeed, to connect in parallel power supply and test inputs of a plurality of integrated circuits and test them simultaneously. For example by connecting only four probes to one of the plurality of integrated circuits (ground, supply voltage and two test mode inputs), it is possible to write all 0's or all 1's and to deselect the entire memory array simultaneously in all integrated circuits on a wafer during the test. This circuit allows to use very simple test equipment and reduces dramatically test times avoiding consequent burn in of packaged devices.

FIG. 1 is an integrated memory circuit according a first embodiment of to the present invention.

FIG. 2 shows part of a silicon wafer on which are integrated a plurality of integrated memory circuits according to the present invention.

FIG. 3 is an integrated memory circuit according to a second embodiment of the present invention.

FIG. 4 is of an integrated memory circuit according to a third embodiment the present invention.

FIG. 5 is a circuit diagram illustrating a first test mode pin circuit according to the present invention.

FIG. 6 is a circuit diagram illustrating a second test mode pin circuit according to the present invention.

FIG. 7 is a schematic diagram of an integrated memory device, in accordance with this invention.

FIG. 8 is an electrical diagram, in schematic form, of an address buffer according to the preferred embodiment of the invention.

FIG. 9 is an electrical diagram, in schematic form, of a data buffer according to the preferred embodiment of the invention.

FIG. 10 is an electrical diagram, in schematic form, of a signal buffer according to the preferred embodiment of the invention.

FIG. 11 is a flow chart of a test method according to the present invention.

FIG. 1 shows an integrated circuit die 132 in its form when present on a wafer of semiconducting material. On the wafer, near the die 132, are present several conducting lines 131 and 141, running in the "scribe line" adjacent to the die, connected with input pads 133, 134, 135 and 137 present on the die 132. Die 132 also includes functional part 146 of the circuit that, according to this example, is a memory circuit, such as a static random access memory (SRAM). A first input pad 133, labeled Vss in figure, is connected to a first conducting line 145 and supplies the Vss voltage to internal circuitry 146. A second 134 and a third 135 input pads, labeled TM0 and TM1 in figure, are connected respectively to a second 144 and to a third 143 conducting lines and to internal circuitry 146. Pads 134 and 135 are also connected each to one input of a NOR gate 136, whose output is connected to control terminal of a p-channel switching transistor 140. Source terminal of transistor 140 is connected to input pad 137, labeled Vcct, which is connected to a fourth conducting line 142. Drain terminal of transistor 140 is connected to a first terminal of a fuse 139, the second terminal of fuse 139 is connected to input pad 138, labeled Vcc, which supplies the Vcc voltage to internal circuitry 146.

The following table shows the possible status of the two test mode input signals TM0 and TM1 and the corresponding conditions of functional circuitry 146:

As can be seen, during the normal operation mode the TM0 and TM1 signals must be maintained low. If at least one of these two signals is high the integrated circuit enters the special test mode operation. There are three different test modes. If TM0 is low and TM1 is high all row and column decoders are selected and the entire array of cells is written with 0's, if TM0 is high and TM1 is low all row and column decoders are selected and the entire array of cells is written with 1's, and finally, if both TM0 and TM1 are high all row and column decoders are deselected.

In operation, the output of NOR gate 136 is at a low level when at least one of the two test mode input signals TM0 and TM1 is high. In this condition the switch transistor 140 is on and the supply voltage Vcct, applied to pad 137 through the conductive line 142, reaches the pad 138 and the internal circuitry 146. If both test mode input signals TM0 and TM1 are low, in normal operation mode, the switch transistor 140 is off and pad 138 is electrically isolated from pad 137. In this condition, in fact, the memory circuit is supplied by the Vcc pad 138 and the Vcct pad 137 must be isolated from the circuitry.

The integrated circuit of FIG. 1 is integrated on a p-type silicon substrate. Normally, a p-type substrate is connected to Vss voltage that corresponds to ground or zero Volts. The substrate type determines which power supply, Vcc or Vss, must be switched by the switch transistor. In fact, during scribing of single dies from the wafer, part of connective lines that run in the "scribe line" may be shorted to the substrate. Is clear that, if for example the p-type substrate is normally connected to Vss, the input pad that in normal operation supplies the circuit with the Vcc voltage must not be shorted to Vss. According to this embodiment of the invention the Vcc pad 138, as can be seen in FIG. 1, is disconnected from the Vcct pad which is connected to conductive line 142, during normal operation by means of the switching transistor 140. For the same reason the normal operation mode of the integrated circuit has been chosen corresponding to low logic levels on both the two test mode input signals TM0 and TM1. This assures that an eventual short circuit between any of the conductive lines and the substrate doesn't jeopardize normal operation of the integrated circuit.

Fuse 139 is eventually blown (for a faulty die), during a brief testing of each die before starting first phase of the parallel test of all dies, to isolate a faulty die from other dies on the wafer. In fact at the beginning each die is first tested individually (or a few at a time) in the test modes to verify they operate correctly with no high current or short circuits. If is found an high current path on the die, the fuse 139, placed in series with the switching transistor, is blown and the circuit integrated on that die is disconnected from other dies on the wafer.

FIG. 2 shows part of silicon wafer 150 on which are integrated a plurality of integrated circuit dies according to the present invention. In FIG. 2, four integrated circuit dies 151, 152, 157 and 158, and conductive lines running horizontally, 155 and 156, and vertically, 153 and 154, between the integrated circuit dies, are shown. These conductive lines are used, during test of integrated circuits at wafer level, to connect in parallel corresponding inputs of a plurality of integrated circuit dies, and are placed in the "scribe line" normally present on wafers between adjacent sides of different dies. When dies are scribed at the end of the test, these conductive lines are automatically eliminated. Through these conductive lines is possible, during the stress test, to control in parallel the operation of all dies by connecting test probes only to one of the plurality of dies present on the wafer and, thus, parallel stressing all dies on the wafer.

In FIG. 3 is shown an integrated circuit die 171 on a wafer and, on the wafer near the die 171, several conducting lines 173 and 174 connected with input pads 175, 176, 177 and 179 that are present on the die 171. Die 171 includes functional circuit 185 which is a memory circuit, such as a static random access memory (SRAM). A first input pad 175, labeled Vcc in figure, is connected to a first conducting line 184 and supplies the Vcc voltage to internal circuitry 185. A second 176 and a third 177 input pads, labeled TM0 and TM1 in figure, are connected respectively to a second 183 and to a third 182 conducting lines and to internal circuitry 185. Pads 176 and 177 are also connected each to one input of a NAND gate 178, whose output is connected to control terminal of an n-channel switching transistor 172. Source terminal of transistor 172 is connected to input pad 179, labeled Vsst, which is connected to a fourth conducting line 181. Drain terminal of transistor 172 is connected to a first terminal of a fuse 181, the second terminal of fuse 181 is connected to input pad 180, labeled Vss, which supplies the Vss voltage to internal circuitry 185.

The integrated circuit of FIG. 3 is integrated on an n-type silicon substrate. Normally, an n-type substrate is connected to Vcc voltage, which is a positive voltage. Therefore in this case the switch transistor 172 switches the power supply Vss. In fact, during scribing of single dies from the wafer, part of connective lines that run in the "scribe line" may be shorted to the substrate which will be powered to Vcc. Therefore, Vss pad 180, as can be seen in FIG. 3, is disconnected, during normal operation by means of the switching transistor 172, from the Vsst pad 179 which is connected to the conductive line 181. For the same reason the normal operation mode of the integrated circuit has been chosen corresponding to high logic levels on both the two test mode input signals TM0 and TM1 as can be seen in the following table that shows the possible status of the two test mode input signals TM0 and TM1 and the corresponding conditions of the internal memory circuit:

This assures that an eventual short circuit between any of the conductive lines and the substrate, occurring during scribing of dies, will not jeopardize normal operation of the integrated circuit.

Fuse 181 is eventually blown (for a faulty die), during a brief testing of each die before starting first phase of the parallel test of all dies, to isolate a faulty die from other dies on the wafer, disconnecting his power supply input Vss from the power supply line Vsst used during the test.

The circuits of FIGS. 1 and 3 apply to substrates which are tied to Vss voltage or to Vcc voltage. In the case where a p-type substrate of an integrated circuit is biased by a charge pump integrated with the circuit to a negative voltage Vbb (normally between -1 Volts and -3 Volts), additional measures must be taken. An integrated circuit die which can be used with this particular substrate biasing is shown in FIG. 4.

FIG. 4 shows an integrated circuit die 211 and, near the die 211, several conducting lines 213 and 214 connected with input pads 215, 216, 219 and 222 present on the die 211. A functional part 230 of the circuit is a memory circuit, a static random access memory (SRAM). A first 215 and a second 216 input pads, labeled TM0 and TM1 in figure, are connected respectively to first (228) and second (227) conducting lines and to internal functional circuitry 230. Pads 215 and 216 are also connected each to one input of a NOR gate 221, whose output is connected to control terminal of a p-channel switching transistor 212 and, by means of an inverter gate 220, to control terminal of an n-channel switching transistor 225. Source terminal of p-channel transistor 212 is connected to input pad 222, labeled Vcct, which is connected to a third conducting line 226. Drain terminal of p-channel transistor 212 is connected to a first terminal of a first fuse 224, the second terminal of fuse 224 is connected to input pad 223, labeled Vcc, which supplies the Vcc voltage to internal circuitry 230. Source terminal of n-channel transistor 225 is connected to input pad 219, labeled Vsst, which is connected to a fourth conducting line 229. Drain terminal of n-channel transistor 225 is connected to a first terminal of a second fuse 218, the second terminal of fuse 218 is connected to input pad 217, labeled Vss, which supplies the Vss voltage to internal circuitry 230.

Fuses 218 and 224 are eventually blown, during a brief testing of each die before starting first phase of the parallel test of all dies, to isolate a faulty die from other dies on the wafer, disconnecting his power supply inputs Vcc and Vss from the power supply lines Vcct and Vsst used during the test.

During normal operation mode, as in the circuit of FIG. 1, the two test mode input signals TM0 and TM1 must be maintained low. If at least one of these two signals is high the integrated circuit enters the special test mode operation.

If at least one of the two test mode input signals TM0 and TM1 is high, the output of NOR gate 221 becomes low. In this condition both switch transistors 212 and 225 are on and the two supply voltages Vcct and Vsst, applied to pads 222 and 219 through the conductive lines 226 and 229, reach the pads 223 and 217 and the internal circuitry 230, allowing for parallel stressing of die 211 along with its neighbors on the wafer.

If both test mode input signals TM0 and TM1 are low, in normal operation mode, both switch transistors 212 and 225 are off. Pad 223, labeled Vcc, is electrically insulated from pad 222 and pad 217, labeled Vss, is electrically insulated from pad 219. In this condition, in fact, the memory circuit is supplied by the Vcc pad 223 and Vss pad 217, and Vcct pad 222 and Vsst pad 217 must be insulated from the circuitry.

If happens, during scribing of single dies on wafer, that part of connective lines are shorted to the p-type substrate biased to negative voltage Vbb, this situation will not jeopardize normal operation of the integrated circuit. In fact Vcc pad 223 and Vss pad 217, as can be seen in FIG. 4, are disconnected, during normal operation, by means of the two switching transistors 212 and 225, from the Vcct and Vsst pads which are connected to conductive lines 226 and 229.

The normal operation mode of the integrated circuit has been chosen corresponding to low logic levels on both the two test mode input signals TM0 and TM1 because, normally, a negative voltage like -1 Volts or -3 Volts on an input of a logic gate is considered by internal circuits of the gate as a low logic level as well as a zero voltage. As such, shorting of lines 182, 183 to the substrate will not affect circuit operation.

As shown in FIG. 5, if necessary, the generic test mode input pad 250, which receives a generic test mode input signal TMx (i.e. either TM0 or TM1), can be disconnected from the connective line 251 blowing a fuse 253. As shown in FIG. 6, pad 241 may be forced inactive by a simple circuit having an inverter gate 240 with a feedback transistor 244 and, for larger security, also by a pull-down transistor driven by a Power On Reset signal (POR, pulses high on power up)), normally present in most memory circuits. This assures that an eventual short circuit between any of the conductive lines and the substrate doesn't jeopardize normal operation of the integrated circuit. The two test mode input pads TM0 and TM1 can also be bonded, during bonding of all pads of the die with the external pins of the package, to ground or to power supply voltage Vcc to be sure that the logic levels on these two input lines are fixed during normal operation of the device.

It is possible also, at the end of the test of the entire wafer, to blow all fuses placed in series with switching transistors on all dies, to be sure that each the test power supply input on each die, extending to the edge of the die, is isolated from the functional circuitry.

FIG. 7 illustrates an example of a memory circuit 110 integrated on an integrated circuit die according to the present invention. Integrated memory circuit 110 is a static random access memory (SRAM), powered by power supply inputs Vcc and Vss, having its memory cells in memory array 111. Memory cells in memory array 111 are arranged in rows and columns, not shown in figure. It should be noted that the designation of rows and columns in memory 110 use the term row to refer to the array direction in which a plurality of memory cells is selected by way of a wordline. The term column is used in this description to refer to the array direction in which one or more of the memory cells in the selected row are selected for read or write access. It is contemplated that such use of the terms rows and columns is consistent with the general understanding in the art.

Memory cells are selected by row and column decoders 123. These decoders are controlled by an internal address bus 112 exiting from address buffer 114. The address buffer 114 receives on its input the external address bus 115 on which are present the address signals A0-An. Data buffer 120 is used to transfer data to and from the memory array through the two-way data busses 113 and 116, during read and write cycles. Two logic signals, chip-enable (CE bar) and write enable (W bar), pass respectively through the buffers 121 and 122. All these buffers 114, 120, 121 and 122, indicated globally by the reference 124 in the figure, are controlled by two external signals, TM0 and TM1. From these two signals it is possible, during testing of the integrated circuit, to enter special test operation modes and to force the outputs of buffers 114 to predetermined logic values. By applying different logic signals to the input terminals TM0 and TM1 it is possible, for example, to select all row and column decoders, and to contemporaneously write predetermined data into the whole array of memory cells. It is also possible, through the same inputs, to deselect all row and column decoders.

FIG. 8 is an electrical diagram illustrating a circuit 10 for an address buffer 114 according to the present invention. First inverter 36 is composed of an n-channel transistor 16 and a p-channel transistor 15 that receives address pad 13 at their gates. Second inverter 20, third inverter 28 and fourth inverter 29 are connected in series between the output of first inverter 36 and a first output signal line 31, labeled A.sub.TRUE. The source terminal of transistor 15 is connected to a pull-up p-channel transistor 14 which is driven by a signal, labeled OE, exiting from the output of an EXOR gate 11. The source terminal of transistor 16 is connected to a pull-down n-channel transistor 17 which is driven by a signal, labeled OD bar, exiting from the output of a NAND gate 12. The two inputs of the EXOR gate 11, connected together with the two inputs of the NAND gate 12, receive two test mode input signals TM0 and TM1 on the input lines 34 and 35. The Node 37, common to the output of the first inverter 36 and to the input of second inverter 20, is connected to a pull-up p-channel transistor 18, driven by the signal OD bar, and to a pull-down n-channel transistor 19, driven by the signal OE.

Pass gate 33 is connected to the output of the second inverter 20. Fifth inverter 27 is connected between the output of pass gate 33 and second output signal line 30, labeled A.sub.COMP. In the preferred embodiment, A.sub.TRUE and A.sub.COMP are input into an address decoder (not shown).

As known in the art, pass gate 33 is composed of an n-channel transistor 25 and a p-channel transistor 26 with a common source and drain. The signal used to control pass gate 33 comes from the output of a NOR gate 23. The signal OD bar is applied to a first input of the NOR gate 23, while the signal OE is applied, by means of a sixth inverting gate 22, to a second input of the NOR gate 23. The output terminal of the NOR gate 23 is connected directly to the gate of the n-channel transistor 25 and, by means of a seventh inverting gate 21, to the gate of the p-channel transistor 26.

The node 38, common to the output of the pass gate 33 and to the input of the fifth inverter 27, is connected to a pull-up p-channel transistor 32, driven by the signal OD bar, and to a pull-down n-channel transistor 24, driven by the signal OE.

As an example of normal operation of the address buffer 10, if both the signals TM0 and TM1 are low, the output of the EXOR gate 11 (signal line labeled OE) is low, while the output of the NAND gate 12 (signal line labeled OD bar), is high. This condition causes both the pull-up transistor 14 and the pull-down transistor 17 to be on, so that the signal present on the address pad 13 is inverted by the first inverter 36 and reaches the node 37. The pull-up and pull-down transistors 18 and 19, connected to the node 37 are both off, so the signal present on node 37 is inverted by the second inverter 20 and reaches pass gate 33. The pass gate 33, driven by the gates 22 and 23, in this condition is on so that the signal present on its input can pass to its output 38. The pull-up and pull-down transistors 32 and 24, connected to the node 38 are both OFF, so the signal present on this node is inverted by the fifth inverter 27 and reaches the second output signal line 30, labeled A.sub.COMP. The signal present on the output of the second inverter 20 reaches also, by means of a third and a fourth inverter 28 and 29, the first output signal line 31, labeled A.sub.TRUE. Inverters 28 and 29 are used to increase drive capacity for the signal A.sub.TRUE. In this condition of normal operation, thus, on the first output signal line 31, labeled A.sub.TRUE is present a signal corresponding to the input signal present on the address pad 13, while on the second output signal line 30, labeled A.sub.COMP, is present a signal complementary to the input signal present on the address pad 13.

For a first example of a test mode operation of the address buffer 10, if the signal TM0 is low and the signal TM1 is high, the output of the EXOR gate 11 (signal line labeled OE) is high, and the output of the NAND gate 12 (signal line labeled OD bar) is high. This condition causes the pull-up transistor 14 to be off and the pull-down transistor 17 to be on. In this condition the inverter 36 does not work as an inverter and its output is forced low by the pull-down transistor 19, which is turned on by the signal OE. The pull-up transistor 18, connected to the node 37 and driven by the signal OD bar, is off. The logic low signal present on node 37 becomes, by means of the three inverters in series 20, 28 and 29, a logic high signal on the first output signal line 31, labeled A.sub.TRUE. In this condition, the pass gate 33, driven by the gates 22 and 23, is off and node 38 is forced low by the pull-down transistor 24 which is turned on by the signal OE. The pull-up transistor 32, connected to the node 38 and driven by the signal OD bar, is OFF. The logic low signal present on node 38 becomes, by means of the inverter 27, a logic high signal on the second output signal line 30, labeled A.sub.COMP. Therefore, in this condition, both outputs A.sub.TRUE and A.sub.COMP of the address buffer are high. This means that if the address decoders, which have A.sub.TRUE and A.sub.COMP as inputs, are composed of NAND gates, and these NAND gates are used to select rows and columns in a memory array, then all of the NAND gates have a low voltage level on their outputs. This causes all of the rows and columns in the memory array to be selected in this first test mode.

The second test mode operation of the address buffer is selected by the signal TM0 high and the signal TM1 low. In this condition both the output of the EXOR gate 11 (signal line labeled OE), and the output of the NAND gate 12 (signal line labeled OD bar), are high. This condition causes the pull-up transistor 14 to be off and the pull-down transistor 17 to be on. The inverter 36 does not work as an inverter and its output is forced low by the pull-down transistor 19 which is turned on by the signal OE. The pull-.up transistor 18, connected to the node 37 and driven by the signal OD bar is off. The logic low signal present on node 37 becomes, by means of the three inverters in series 20, 28 and 29, a logic high signal on the first output signal line 31, labeled A.sub.TRUE. The pass gate 33, driven by the gates 22 and 23, is off and node 38 is forced low by the pull-down transistor 24 which is turned on by the signal OE. The pull-up transistor 32, connected to the node 38 and driven by the signal OD bar is off. The logic low signal present on node 38 becomes, by means of the inverter 27, a logic high signal on the second output signal line 30, labeled A.sub.COMP. Therefore, also in this second test mode condition, both outputs A.sub.TRUE and A.sub.COMP of the address buffer are high. This means that if the address decoders, which have A.sub.TRUE and A.sub.COMP as inputs, are composed of NAND gates, and these NAND gates are used to select rows and columns in a memory array, then all of the NAND gates have a low voltage level on their outputs. This causes all of the rows and columns in the memory array to also be selected in this second test mode.

The third test mode operation of the address buffer is selected by both signals TM0 and TM1 high. In this condition the output of the EXOR gate 11 (signal line labeled OE) is low, and the output of the NAND gate 12 (signal line labeled OD bar) is low. This condition causes the pull-down transistor 17 to be off. The inverter 36 does not work as an inverter and its output is forced high by the pull-up transistor 18 which is turned on by the signal OD bar. The pull-down transistor 19, connected to the node 37 and driven by the signal OE is off. The logic high signal present on node 37 becomes, by means of the three inverters in series 20, 28 and 29, a logic low signal on the first output signal line 31, labeled A.sub.TRUE. The pass gate 33, driven by the gates 22 and 23, is off and node 38 is forced high by the pull-up transistor 32 which is turned on by the signal OD bar. The pull-down transistor 24, connected to the node 38 and driven by the signal OE bar is off. The logic high signal present on node 38 becomes, by means of the inverter 27, a logic low signal on the second output signal line 30, labeled A.sub.COMP. Therefore, in this third test mode condition, both outputs A.sub.TRUE and A.sub.COMP of the address buffer are low. This means that if the address decoders, which have A.sub.TRUE and A.sub.COMP as inputs, are composed of NAND gates, and these NAND gates are used to select rows and columns in a memory array, then all of the NAND gates have a high voltage level on their outputs. This causes all of the rows and columns in the memory array to be deselected in this third test mode.

FIG. 9 is an electrical diagram illustrating a circuit 40 for a data buffer according to the present invention. A first inverter 63, composed of an n-channel transistor 46 and a p-channel transistor 45 with common gates and a second inverter 50 are connected in series between a data pad 43 and an output signal line 51, labeled DATA. The source terminal of transistor 45 is connected to a pull-up p-channel transistor 44 which is driven by a signal, labeled OE, exiting from the output of a first AND gate 41. The source terminal of transistor 46 is connected to a pull-down n-channel transistor 47 which is driven by a signal, labeled OD bar, exiting from the output of a NAND gate 42. A first input of the first AND gate 41, connected together with a first input of the NAND gate 42, receive an output signal from a first OR gate 52, whose inputs are connected to the input lines 62 and 54 on which are present the two "Test mode" input signals TM0 and TM1. A second input of the first AND gate 41 is connected, by means of a third inverting gate 53, to the input line 62 on which is present the input signal TM0. A second input of the NAND gate 42 is connected to the output of a second OR gate 56, whose first input is connected, by means of a fourth inverter gate 57, to the input line 54 on which is present the input signal TM1. A second input of the second OR gate 56 is connected to the output of a second AND gate 60, whose inputs are connected to the input lines 62 and 54 on which are present the two input signals TM0 and TM1.

The node 64, common to the output of the first inverter 63 and to the input of second inverter 50, is connected to a pull-up p-channel transistor 48, driven by the signal OD bar, and to a pull-down n-channel transistor 49, driven by the signal OE.

As an example of normal operation of the data buffer 40, if both the signals TM0 and TM1 are low, the logic gates 52, 53 and 41 generate a logic signal low on the signal line labeled OE, and the logic gates 56, 57, 60 and 42 generate a logic signal high on the signal line labeled OD bar. This condition causes both the pull-up transistor 44 and the pull-down transistor 47 to be on, so that the signal present on the data pad 43 is inverted by the first inverter 63 and reaches the node 64. The pull-up and pull-down transistors 48 and 49, connected to the node 64, are both off, so the signal present on node 64 is inverted by the second inverter 50 and reaches the output signal line 51, labeled DATA. In this condition of normal operation, a signal corresponding to the input signal present on the data pad 43 is present on the output signal line 51, labeled DATA.

For a first example of a test mode operation of the data buffer 40, if the signal TM0 is low and the signal TM1 is high, the logic gates 52, 53 and 41 generate a logic signal high on the signal line labeled OE, and the logic gates 56, 57, 60 and 42 generate a logic signal high on the signal line labeled OD bar. This condition causes the pull-up transistor 44 to be off and the pull-down transistor 47 to be on. In this condition the inverter 63 does not work as an inverter and its output, on node 64, is forced low by the pull-down transistor 49 which is turned on by the signal OE. The pull-up transistor 48, connected to the node 64 and driven by the signal OD bar is off. The logic low signal present on node 64 becomes, by means of the second inverter 50, a logic high signal on the output signal line 51, labeled DATA. In this first test mode condition is present a logic signal high on the output signal line 51, labeled DATA. Since all rows and columns are selected in this first test mode, as described above, a "one" data state is written into all memory cells currently selected by address decoders in this first test mode.

As noted above, the second test mode operation of the data buffer 40 is selected by the signal TM0 high and the signal TM1 low. In this condition the logic gates 52, 53 and 41 generate a logic signal low on the signal line labeled OE, and the logic gates 56, 57, 60 and 42 generate a logic signal low on the signal line labeled OD bar. This condition causes the pull-up transistor 44 to be on and the pull-down transistor 47 to be off. In this condition the inverter 63 does not work as an inverter and its output, on node 64, is forced high by the pull-up transistor 48 which is turned on by the signal OD bar. The pull-down transistor 49, connected to the node 64 and driven by the signal OE, is off. The logic high signal present on node 64 becomes, by means of the second inverter 50, a logic low signal on the output signal line 51, labeled DATA. In this second test mode condition, thus, on the output signal line 51, labeled DATA, is present a logic signal low. Since all rows and columns are selected in this second test mode, a "zero" is written into all memory cells currently selected by address decoders.

The third test mode operation of the data buffer 40 is selected by both signals TM0 and TM1 high. In this condition the logic gates 52, 53 and 41 generate a logic signal low on the signal line labeled OE, and the logic gates 56, 57, 60 and 42 generate a logic signal low on the signal line labeled OD bar. The data buffer is in a condition equivalent to the second test mode operation and the output signal line 51, labeled DATA, is forced to a low level. Therefore in this third test operation mode all rows and columns are deselected by address decoders precluding data from being written into any memory cells and the buffer is maintained in a defined state.

FIG. 10 is an electrical diagram illustrating a circuit 70 for a signal buffer according to the present invention. This signal buffer is used, according to the present invention, to buffer signals W bar and CE bar used respectively to enable write cycles and to enable chip operation, as these two signals are both active low and require the same kind of buffer. Therefore, while in FIG. 10 the input pad and the output signal line are labeled "W bar", the same circuit can be used for the Chip Enable signal "CE bar".

The signal buffer circuit 70 has a first inverter 85, composed of an n-channel transistor 76 and a p-channel transistor 75 with common gates, and a second inverter 80 connected in series between an input pad 73 and an output signal line 83, labeled W bar. The source terminal of transistor 75 is connected to a pull-up p-channel transistor 74 which is driven by a signal, labeled OE, exiting from the output of an AND gate 71. The source terminal of transistor 76 is connected to a pull-down n-channel transistor 77 which is driven by a signal, labeled OD bar, exiting from the output of a third inverting gate 86, whose input is connected to the output of an EXOR gate 72. The two inputs of the AND gate 71, connected together with the two inputs of the EXOR gate 72, receive the two test mode input signals TM0 and TM1 on the input lines 81 and 82. The node 84, common to the output of the first inverter 85 and to the input of second inverter 80, is connected to a pull-up p-channel transistor 78, is driven by the signal OD bar, and is connected to a pull-down n-channel transistor 79, driven by the signal OE.

As an example of normal operation of the signal buffer 70 of FIG. 10, if both the signals TM0 and TM1 are low, the AND gate 71 generates a logic signal low on the signal line labeled OE, and the EXOR gate 72 generates a logic signal low on its output which, inverted by the third inverter 86, becomes a logic signal high on the signal line labeled OD bar. This condition causes both the pull-up transistor 74 and the pull-down transistor 77 to be on, so that the signal present on the signal pad 73 is inverted by the first inverter 85 and reaches the node 84. The pull-up and pull-down transistors 78 and 79, connected to the node 84, are both off, so the signal present on node 84 is inverted by the second inverter 80 and reaches the output signal line 83, labeled W bar. In this condition of normal operation, thus, a signal corresponding to the input signal present on the signal pad 73 is present on the output signal line 83, labeled W bar.

For a first example of a test mode operation of the signal buffer 70, if the signal TM0 is low and the signal TM1 is high, the AND gate 71 generates a logic signal low on the signal line labeled OE, and the EXOR gate 72 generates a logic signal high on its output which, inverted by the third inverter 86, becomes a logic signal low on the signal line labeled OD bar. This condition causes the pull-up transistor 74 to be on and the pull-down transistor 77 to be off. In this condition the inverter 85 does not work as an inverter and its output, on node 84, is forced high by the pull-up transistor 78 which is turned on by the signal OD bar. The pull-down transistor 79, connected to the node 84 and driven by the signal OE, is off. The logic high signal present on node 84 becomes, by means of the second inverter 80, a logic low signal on the output signal line 83, labeled W bar. In this first test mode condition, thus, a logic signal low is present on the output signal line 83, labeled W bar. Since all rows and columns are selected in this mode, and since data buffer 40 is driving a logic "one" state, a "one" data state is written into all memory cells selected by address decoders in this first test mode.

The second test mode operation of the data buffer 40 is selected by the signal TM0 high and the signal TM1 low. In this condition the AND gate 71 generates a logic signal low on the signal line labeled OE, and the EXOR gate 72 generates a logic signal low on the signal line labeled OD bar. The signal buffer is in a condition equivalent to the first test mode operation and the output signal line 83, labeled W bar, is forced to a low level. This means that memory is enabled to start a "write cycle". In this test mode, as described, all rows and columns are selected, and data buffer 40 is driving a "zero" level, to write a "zero" into all cells simultaneously.

The third test mode operation of the signal buffer 40 is selected by both signals TM0 and TM1 high. In this condition the AND gate 71 generates a logic signal high on the signal line labeled OE, and the EXOR gate 72 generates a logic signal high on the signal line labeled OD bar. This condition causes the pull-up transistor 74 to be off and the pull-down transistor 77 to be on. In this condition the inverter 85 does not work as an inverter and its output, on node 84, is forced low by the pull-down transistor 79, which is turned on by the signal OE. The pull-up transistor 78, connected to the node 84 and driven by the signal OD bar, is off. The logic low signal present on node 84 becomes, by means of the second inverter 80, a logic high signal on the output signal line 83, labeled W bar. In this third test mode condition, thus, on the output signal line 83, labeled W bar, is present a logic signal high that disables writing in all memory array.

A method for stress testing a plurality of dies on a single silicon wafer each of the plurality integrating a memory circuit that incorporates the above described buffer circuits will now be described. Each die of the wafer is first tested briefly individually (or a few at a time) in the test modes to verify they operate correctly with no high current or short circuits. If a high current path is found on one die, the fuse placed in series with the switching transistor is blown and the circuit integrated on that die is disconnected from other dies on the wafer.

For a p-type substrate die, such as the die shown in FIG. 1, the fuse 139 is blown which disconnects the power supply input Vcc of the functional circuitry 146 from the power supply line Vcct used during the test.

Analogously, for an n-type substrate die, such as the die shown in FIG. 3, the fuse 181 is blown which disconnects the power supply input Vss of the functional circuitry 185 from the power supply line Vsst used during the test.

If the p-type substrate of the die is biased to a negative voltage, such as the die of FIG. 4, both the fuses 218 and 224 are blown which disconnect both power supply input Vcc and Vss of the functional circuitry 230 from the power supply lines Vcct and Vsst used during the test.

In fact, originally, correspondent pads of all dies on the wafer are parallel connected by conductive lines running in the "scribe lines" between dies and controlling input pads on one die is possible to control circuits integrated on all dies. In the same way, parallel connecting correspondent pads of single rows or columns of dies or single portion or quadrant of the wafer, it is possible to control circuits integrated on all that portion of the wafer.

FIG. 11 shows a flow chart of the stress test method. The test is described referring to a single memory device integrated on a single die because all other dies on the wafer are controlled in parallel simultaneously in the same way. According to the preferred embodiment of the invention, one stress test cycle consists of three test phases, each test phase defined by the combination of test mode control signals TM0 and TM1 and corresponding to one test operation mode explained above and shown in the following table:

The first phase of the test starts by powering up the integrated memory circuit to a stress voltage, e.g. 7 Volt, which is higher than the normal operation voltage of the device. This power up is done while the first test mode control signal TM0 is maintained low and the second test mode control signal TM1 is maintained high. As has been explained above, this combination of test mode control signals forces all outputs of address buffers high to select of all rows and columns in the array of memory cells, forces all outputs of data buffers low, and forces outputs of the two signal buffers for signals "write enable" and "chip enable" low. This situation causes a parallel writing of all memory cells in the array with logic "zeroes". This operation does not require, as in conventional devices, separate control of all address, data and control pads, but only requires control of two signals, TM0 and TM1. This means an effective reduction of test time and of complexity of test probes and test equipment. During this first step, shown in FIG. 5 as process 101, the bitline loads are disabled.

The next step, shown by process 102, is to heat up the integrated memory circuit to a high temperature, e.g. 200.degree.-250.degree. C. and to pause for a certain amount of time, seconds up to several minutes. The stress condition, maintained on the memory array during process 102, allows test for ionic contamination or charge trap sites. This is allowed by the greater mobility of ions at high temperatures and their tendency to drift (or be attracted) based on an electric field.

Then, in process 103, exited test mode, the device is cooled down and, in process 104, is exercised by writing and reading each memory cell individually, or by bytes or words, to identify any and all failures in the memory array. This particular process 104 cannot be made simultaneously on all dies on the wafer and is made singularly, on a different tester station, on each single die. If this verify step of process 105 fails, the device is discarded (or indicated as a failure on the wafer) and the stress test ends. If, otherwise, the verify step is passed, the device is powered down in process 106, ending the first phase of the stress test.

The second phase of the stress test starts by again powering up the integrated memory circuit to the same stress voltage while the first test mode control signal TM0 is maintained high and the second test mode control signal TM1 is maintained low. As has been explained above, this combination of test mode control signals forces all outputs of address buffers high to select all rows and columns in the array of memory cells, forces all outputs of data buffers high, and forces outputs of the two signal buffers for signals "write enable" and "chip enable" low. This situation causes a parallel writing of all memory cells in the array with logic "ones". This operation, as in first stress test phase, does not require controlling all address, data and control pads, but requires control of only the two test mode signals TM0 and TM1. During this first step, is shown in FIG. 5 as process 201, the bitline loads are disabled.

The next step, shown as process 202, is to heat up the integrated memory circuit to a high temperature, e.g. 200.degree.-250.degree. C. and to pause for a certain amount of time, seconds up to several minutes. The stress condition, maintained on the memory array during process 202, allows test for ionic contamination or charge trap sites. This is allowed by the greater mobility of ions at high temperatures and their tendency to drift (or be attracted) based on an electric field.

Then, in process 203, the device is cooled down and, in process 204, exited test mode, is exercised by writing and reading each memory cell individually, or by bytes or words, to determining if any of the memory cells in the array are now failures. This verify process 204 cannot be made simultaneously on all dies on the wafer and is made singularly, on a different tester station, on each single die. If this verify step of process 205 fails, the device is identified as a failure and the stress test ends. If, otherwise, the verify step is passed, the device is powered down in step 206 ending the second phase of the stress test.

The third and last phase of the stress test starts by again powering up the integrated memory circuit to the same stress voltage, while the first and the second test mode control signals TM0 and TM1 are maintained high. As explained above, this combination of test mode control signals forces all outputs of address buffers low to deselect of all rows and columns in the array of memory cells, forces all outputs of data buffers low, and forces outputs of the two signal buffers for signals "write enable" and "chip enable" high. In this situation all rows and columns are deselected, writing is disabled, and the "high" chip enable signal causes the disabling of most circuits inside the integrated memory circuit. This operation, as well as in other stress test phases, does not require controlling of all address, data and control pads, but of only the two test mode signals TM0 and TM1. This first step is shown in FIG. 11 as process 301.

Next step, shown as process 302, is to heat up the integrated memory circuit to a high temperature, i.e. 200.degree.-250.degree. C. and to pause for a certain amount of time, seconds up to several minutes. The stress condition, maintained on the memory array during process 302, allows test for ionic contamination or charge trap sites. This is allowed by the greater mobility of ions at high temperatures and their tendency to drift (or be attracted) based on an electric field.

Then, in process 303, the device is cooled down and, in process 304, exited test mode, is exercised by writing and reading each memory cell individually, or by bytes or words, to determine if any portion of the memory is now inoperable. This verify process 304 cannot be made simultaneously on all dies on the wafer and is made singularly, on a different tester station, on each single die. If this verify step of process 305, fails the device is identified as a failure and the stress test ends. If, otherwise, the verify step is passed, the device is powered down in process 306, ending the stress test.

In order to enhance oxide stressing, it is possible to perform an additional oxide stress step, after the steps of cooling down the device (processes 103, 203 and 303), and before exiting the test mode and exercising memory cells individually (processes 104, 204 and 304). This step can be performed in all three described phases, or only in one or two of them, performed by cooling the device to a medium temperature, for example 80.degree. C., raising the supply voltage to a higher stress voltage, for example 9 Volts, and maintaining these conditions for a limited time, for example one second. This step is believed to improve the stress of all oxides in the integrated circuit.

The above described method for parallel stress testing of integrated circuit dies, together with the above described circuits, provides a method for detecting latent defects within the memory array as well as within decoders and periphery circuits. DC conditions of the test, compared to traditional dynamic testing at elevated temperature, reduce latchup phenomena and allows test for ionic contamination or charge trap sites. This is allowed by the greater mobility of ions at high temperatures and their tendency to drift (or be attracted) based on a constant electric field.

Furthermore, the stress test described above saves time because it performs, at wafer level, a complete stress test that would otherwise be done, singularly, in a burn in on packaged devices. All of the stress test can be performed in less than several minutes, or even few seconds, instead of the several hours (such as 96 hours for a 1 Megabyte SRAM) necessary for a singular conventional burn in test. Finally, the test equipment is also simplified because, during the test, all the integrated circuit dies are controlled by only four pads, two for power supply and two for controlling different test operation modes.

Although the method has been described by applying the three phases in such order, writing all zeroes in the memory array, writing all ones in the memory array and then deselect all memory array, other combination or subsets of the three test phases may be used. For example the sequencing of the test phase may be altered as follows, first deselect all memory array and then write into the memory array all zeroes and then all ones.

In addition, the stress time, the stress voltage applied to the integrated circuit and the temperature are related parameters, and they can vary between certain limits. For example longer time at higher voltage may allow reduced temperature or higher voltage at higher temperature may allow reduced time.

Finally, the verification processes may all be done at the end of all three test phases, so that these three phases can be done on the same simple test station. Indeed the verification process has to be done on a different tester station that permits to read and write single cells, or bytes or words, in the memory arrays of single dies, controlling any address and data signals in normal operation mode.

In other words the method described above may be changed to meet the needs of the user.

While the invention has been particularly shown and described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention.