Power Supply Current Detectability of SRAM Defects Jian Liu and Rafic Makki Department of Electrical Engineering University of North Carolina at Charlotte, Charlotte, NC 28223
Abstract We investigate correlations between SRAM cell defects and power supply current including ZDDQ (peak value of quiescent power supply current) and iDDT (transient power supply current). Our results show that the power supply current can be used to detect cell shorts, cell opens, and disturb-type pattem sensitivity. We also investigate the effect of total current leakage in the power supply, which is proportional to SRAM size, on the power supply current detectability of SRAM cell defects. We present limits (obtained from simulation) on current detectability of SRAM cell defects as a function of normal power supply current leakage. Keywords: transient power supply current, quiescent power supply current, physical defects, large circuit sizes
I. Introduction At the 1981 International Test Conference Levi showed that fully CMOS (FCMOS) circuits can be tested, in part, by measuring ZDDQ. Normal ZDDQ levels in a FCMOS circuit are mostly due to current leakages. Elevated IDDQ levels could indicate circuit defects such as gate-oxide shorts. This is a significant result because ZDDQ is directly observable and yet can indicate defects anywhere in the circuit. Since that time the test industry has recognized the advantages of ZDDQ and a wealth of ZDDQ research results have been compiled [11-[29]. Researchers have developed sensors for built-in monitoring of ZDDQ [15]-[21], proposed new test generation techniques [21][%I, studied the use of scan techniques to enhance controllability for ZDDQ testing [251, experimented with ZDDQ. in production testing 1261, utilized ZDDQ for fault locatton [27], and reported on testing static RAMS with IDDQ [28][29]. The test industry is in the process of developing standards for off-chip ZDDQ sensors as was presented at the recent 1994 International Test Conference [30]. Recently, iDDT has been proposed as an additional testability measure for detecting SRAM defects including cell disturbs [31]. The iDDT provides a window of observability into the internal switching characteristics of a circuit. Every time a circuit switches, it is expected to produce an ~DDT pulse whose peak level is appreciably greater than IDDQ. 0-8186-7129-7/95$4.00 0 1995 IEEE The Fourth Asian Test Symposium
The iDDT pulse peak level is due to two components; direct transient path between power and ground, and path between the output of the switching gate (gates) and either power or ground (depending on whether the circuit is switching from a high state to a low state or from a low state to a high state). Defects such as opens may prevent the gate from having a normal conducting path between power and ground during switching which can appreciably change the level of iDDp A power tree distribution technique was presented in order to isolate normal current transients from those which result from defects. Experimental results on a small size SRAM (8K) indicate that when combined together, ZDDQ and ZDDT present a formidable test that can reduce the number of escaping defects [31] 4331 (100% coverage of both read and write single cell disturbs was obtained) using test lengths of 5n where n is the number of cells.
In this paper, we investigate correlations between SRAM opens, shorts, and cell disturbs as a function of circuit size. We attempt at qualifying the power supply current detectability of these defects as a function of total current leakage which is a function of SRAM size. Large SRAMs are typically divided into a number of array islands. The results presented in this paper can be used as estimates on the size of arrays that can practically be tested with power supply current.
II.Background In this section we present simple examples on the power supply current detectability of SRAM defects. Throughout this paper, the following notation shall be employed. &i denotes the complete supply current ZDDQ: denotes the level of the quiescent supply current ZDDY denotes the peak value of the transient supply current ~DDT:denotes the complete transient supply current
II. 1Z D ~ Example e Consider the SRAM cell of Figure 1-b and assume
367
Authorized licensed use limited to: University of Central Florida. Downloaded on October 25, 2008 at 23:37 from IEEE Xplore. Restrictions apply.
B
that there is a resistive bridge between the gate line of transistor m4 and the data line d connecting the gate of transistor m4 to Vdd and the Bit line to GND results in a path via the bridge between Vdd and GND. Such a path may increase the level of JDDQ. In using IDDQ to detect the fault, there are several issues. These issues include: the level of normal leakage into the power supply, the level of current noise in the power supply, and the smallest current resulting from a defect which can be detected by a sensor [34]. If the level of IDDQresulting from a defect is outside the design range (appreciably higher than the normal; ZDDQ), the defect can be detected by a sensor. However, if the level of ZDDQ resulting from a defect is not a p p i a bly different from the normal leakage level (lies within the normal noise level or is smaller than the smallest defective that can be detected by a sensor), the defect may not be detected by testing. The remedy would be to partition the circuit into segments that individually have lower normal leakage levels. These issues affect the selection of a pass/fail value for ZDDQ limit. The ZDDQ limit settings can significantly vary from circuit to circuit.
the switching of ce112. Recently, Su et. al. [31] have pro-
(4
GND (b)
Figure 1. Normal S R A M Cell Structure and Model
posed a method for power supply distribution that uses an additional test Vdd line for isolating normal iDDT pulses from those resulting from defects. For convenience, Figure 3 shows an example of this power tree partitioning technique.
II.2 iDDTExample Assume that the drain of transistor m l is open in Figure 1-b and that the cell is initially set to logic 1. If the cell is intended to switch to logic 0 by applying a 0 to the bit line and activating the word line, a normal iDDT pulse would result. However, in the presence of the open, the actual ZDDT level may be appreciably smaller than the expected value. A current sensor can detect this difference by allowing the iDDT pulse to charge a capacitor and using an analog comparator to compare the resulting voltage to a reference vdue. Now consider the cell pair of Figure 2. If one performs a write operation to celll involving switching of its state, only celll is expected to switch and cell2 should be unaffected. However, in the presence of a cell disturb, the state of cell2 may also switch as a result of writing to celll. This results in a transient pulse due to
Figure 2. Cell Disturb Fault
PassiFail Vdd
.
System GND
,
Figure 3. Power Tree distribution for iDDT Observability
368
Authorized licensed use limited to: University of Central Florida. Downloaded on October 25, 2008 at 23:37 from IEEE Xplore. Restrictions apply.
The SRAM utilizes one Vdd line (Vdd2) and one GND line (GND2) dedicated for viewing abnormal ~DDT pulses. In this scheme, if a particular cell is addressed, say cell(l,l), its Vdd and GND lines get connected to the normal System power and GND lines. All other cells are connected to either Vdd2, GND2, or both. The test power and GND lines are connected to ~ D D Tsensors. If cell (1,l) disturbs any other cell in the array, an abnormal iDDT pulse will be sensed on either Vdd2 or GND2. The circuitry required to perform the power tree distribution consists of a pair of transistors per array column and a pair per array row [31]. An iDDT sensor involves using the transient pulse to charge a capacitor and comparing the resulting charge to a reference value. In some respects, the design of the iDDT sensor is simpler than that of ZDDQ because, in the latter case, one has to not allow the transient pulse which precedes the quiescent current to disrupt the quiescent current measurement. The issues involved in ~DDTtesting include: the smallest pulse that can be detected by a sensor, the normal system noise levels, the normal amount of leakages, and the amount of capacitance on the power supply connected to the sensor. In the case of open faults, an important issue is the open fault voltage because that could affect the excitation of the power-GND path. Other issues include the effects of power tree partitioning in terms of performance degradation as addressed in [321.
are detected between the two sets of write operations. However, practical size SRAh43 have much higher leakage currents and much bigger capacitances on the power supply connected to the current sensors. Both of these factors can influence the current detectability of defects as was discussed in Section 11. The effect of these parameters is discussed in more detail in Section V.
IV. Open Defects We considered five different open defects. An important consideration is the voltage on the open which can be due to parasitic capacitances. Table 2 shows the effect of the open voltage on both ZDDQ and ZDDT for a double floating gate defect. The open voltage was varied from 0.5 V to 4.5V. The ZDDQ and ZDDT levels were recorded for a 0 to 1 write operation and 1 to 0 write operation. For the write1 operation, the faulty current values approach the good current values as the voltage in the open approaches 4.5 V, as is expected. However, for the write-0 Operation, the faulty current responses increase as the voltage in the open increases because the write operation results in a 1 on db (Figure 1) which in turn tries to pull the open to 0. As a result, the open is being pulled high and low simultaneously giving rise to a Vdd to GND short which becomes stronger as the open voltage is increased.
IaaaaBs
In the case of an open voltage of 0.5 V, two of the opens were detected with the write-l operation and one with the write-0 operation. Two of the opens result in a AZDDQof approximately 1 pA (for the write-0 operation) with normal leakage of 6.6 pA. This could result in a detect depending on the levels of normal system noise margins and the sensor characteristics. An appreciable ZDDT difference results for all opens in the case of the write 0 and write 1 operation. In the case of the double floating gate defect, although we see an appreciable value of NDDT(better than 50% change from normal ZDDT), the resulting Test AVdd, which is a better measure, is less than 50% of the normal value. Thus, the open may or may not be detected depending upon variables such as normal system noise and accuracy of the current sensor.
III. Short Defects In this section, we investigate the effect of shorts including those that can cause faults on power supply current levels. We considered five different resistive shorts. Table 1 gives the simulation results for a 0 to 1 write operation. Similar results were obtained for a 1 to 0 write operation. The row labelled “good” represents the good circuit response. The data were obtained by using a standard SRAM cell layout (from Cascade Logic tool set) and performing Spice simulation using the HP 1.2 pm process parameters. The LUDDQ/ AZDDT column refers to the change in quiescent/transient current between the good and faulty circuits. In the case of ~DDTa good measure is the effect of the current (area of iDDT pulse) in terms of charging the capacitor on the supply line connected to the tester. This is quantified in the Test Vdd and Test AVdd columns. Establishing detectability percentages requires setting current limits to be used by the sensors. In the case of ZDDQ, all shorts provide ZDDQ levels which are significantly different than the good circuit levels when both write operations (0 to 1 and 1 to 0) are performed. Thus these defects should be easily detected by ZDDQ for any reasonable current limit. In the case of iDDT, if we assume a detect for up to a 1 V change in voltage, all the shorts
@eh ODen Voltage An open voltage of 4 . W is probably unrealistic but gives a sense of what happens to the current levels in extreme situations. However, as we saw from Table 2, the current detectability profile of the open does not significantly change as the open voltage goes beyond 2.5 V. We found that three of the opens may be detected with ZDDe if the sensor can delineate a 1 pA change for normal operat-
369
Authorized licensed use limited to: University of Central Florida. Downloaded on October 25, 2008 at 23:37 from IEEE Xplore. Restrictions apply.
Short4 Short5
1601.889 294.669
I 1595.278 I
903.7 174.4
288.058
738.0 8.7
4951.6 2759.3
3499.6 1307.3
Table 2: Effect of Open Voltage for a Double Floating Gate Defect. ~~
Cell Good
Initial Voltage (V)
N/A
Defect
0.5
Defect
1.o
Defect
1.5
Defect
2.0
Defect
2.5
Defect
3.5
Defect
4.5
Test 0-->1 1 -->0 0--> 1 1 -->0 0-->1 1--> 0 0--> 1 1--> 0 0-->1 1--> 0 0--> 1 1 -->0 0-->1 1 --> 0 0--> 1 1--> 0
Table 3: InDeand i
~
~
~-
ZDDQ @A)
6.611 6.612 139.988 11.672 111.321 41.939 86.789 165.357 64.341 373.385 44.471 594.439 13.280 776.318 6.774 765.904
NDDQ (PA)
ZDDT
-i"-
165.7 168.4 252.9 4.8 226.5 35.3 83.9 141.6 54.3 131.6 35.2 120.6 4.1 107.3 1.3 110.9
0.OOO 133.387 5.060 104.710 35.327 80.178 158.745 57.730 366.773 37.860 587.827 6.669 769.706 0.163 759.292
(km
MDDT @A)
1 T . o 0.0 87.2 -163.6 60.8 -133.1 -82.7 -26.8 -111.4 -36.8 -130.5 -47.8 -161.6 -61.1 164.0 -57.5
Test Vdd (mV) 1480.6 1486.9 207 1.3 1546.7 1847.4 1725.7 1635.6 2272.2 1467.3 2105.8 1387.1 2430.3 1353.2 2594.4 1350.4 2694.1
~ Current n ~ of the Five Types of Cell-Cell Shorts.
Authorized licensed use limited to: University of Central Florida. Downloaded on October 25, 2008 at 23:37 from IEEE Xplore. Restrictions apply.
i
Test AVdd (my) 0.0 0.0 590.7 59.8 366.8 238.8 155.0 785.3 -13.3 618.9 -93.5 943.4 -127.4 1107.6 -130.2 1213.5
ing levels of 6 PA. Two of the opens are easily detected by Z D D ~ with the write-0 operation. In iDDT testing, all opens are detected except one open drain because the high voltage on the open drain representing the output of the inverter affects its switching characteristics only to a very small extent.
10, 000 cells. Since Spice simulations run times are prohibitive, the size of the SRAM was emulated by adding leakage into the power supply and connecting a capacitor on the power bus connected to the sensor. The SRAM leakage and capacitance values of a single cell were used to compute the estimates for the larger SRAM sizes. Figure 4 and 5 show a summary of the results. The Figures indicate the percentage of change in current between the good and faulty circuits for different values of normal leakage currents. This data can be used to help determine the maximum normal leakage current for a given sensor measurement precision. In turn, this can help in power tree partitioning and/or determination of number of sensors required.
V. Large Circuit Effects In this section, we examine the current detectability of SRAM shorts and opens as the level of normal current leakage gets large and the value of the capacitance on the test power bus connected to the tester gets large. We chose two faults for this analysis: resistive bridge and double floating gate. The SRAM size was varied from 10 cells to
Figure 5.The Current Percentage Change vs the Leakage Current for bndging Defect.
371
Authorized licensed use limited to: University of Central Florida. Downloaded on October 25, 2008 at 23:37 from IEEE Xplore. Restrictions apply.
598-607. [8] S . McEuen, “IDDQ Benefits,” Proc. of IEEE VLSI Test Symposium, 1991, pp. 285-290. [9] S . McEuen, “Why IDDQ?,” Proc. of International Test Conference, 1990, pp. 285-290 [lo] T. Storey and W. Maly, “CMOS Bridging Fault Detection,” Proc. of International Test Conference, 1990, pp. 842-851. [ll] K. Baker and B. Verhelst, “IDDQtesting because Zero defects isn’t enough: A Philips Perspective,”Proc. of International Test Conference, 1990, pp. 253-254. [12] C. Hawkins, J. Soden, R. Fritzemeier and L. Horing, “Quiescent Power Supply Current Measurement for CMOS IC Defect Detection,” IEEE Tran. on Industrial Electronics, Vol. 36, No. 2. May 1989,pp. 211-218. [13] R Fritzemeier, C. Hawkins, and M. Soden, “CMOS IC Fault Models, Physical Defect Coverage, and IDDQ Testing,” Proc. of IEEE 1991 Custom Integrated Circuits Conference, pp. 13.1.1- 13.1.8. [14]R Rodriguez-Montanes, J. Segura, V. Champac, J. Figueras, and J. Rubio, “Current vs. Logic Testing of Gate Oxide Short, Floating Gate and Bridging Failures in CMOS,” Proc. of International Test Conference, 1991, pp. 510-519. [15] A. Welbers, B. Verhelst, and E. Seevinck, and K. Baker, “A Built-in CMOS Idd Quiescent Monitor Circuit,” Proc.of IEEE VLSI Circuit Symposium. [16] W. Maly, and P. Nigh, “Built-in Current Testing-A Feasibility Study,” Proc. of International Conference on Computer-Aided Design, November 1988, pp. 340-343. [17] P.J. Nigh, “Built-in Current Testing,” SRC TechnicaE Report 730083, July 1990. E181 P. Nigh and W. Maly, “A Self-Testing ALU Using Built-In Current Sensing,” Proc. of IEEE 1989 Custom Integrated Circuit Conference, pp. 22.1.122.1.4. [19] D. Feltham, P. Nigh, L. Carley, and W. Maly, “Current Sensing Built-In Testing of CMOS Circuits,’’ Proc. of ICCD 1988, pp. 454-457. [20] M. Patyra and W. Maly, “Circuit Design for Built-In Current Testing,” Proc. of IEEE I991 Custom Integrated Circuits Conference, pp. 13.4.1-13.4.5. [21] P. Nigh, and W. Maly, “Test Generation for Current Testing,” IEEE Design and Testing of Computers, Feb. 1990,pp. 26-38. [22] W. Mao, R. Gulati, D. Goel, and M. Ciletti, “QUIETEST A Quiescent Current Testing Methodology for Detecting Leakage Faults,” Proc. of
VI. Disturb Defects In this section, we investigate the effect of disturb faults on IDDQ and iDDT. We considered five bridging defects between two SRAM cells. The test results are given in table 3. It was found that defects 2,3, and 4 cause cell disturbs when both cells are initialized to 0 and a write-1 operation is performed on the top cell.We did not employ the testable SRAM structure of Figure 3 which would have made the defects easily detected by iDDT since no iDDT pulse would have been expected on the supply line of cell2 if only cell1 is written to. The test results Show iDDT to detect dl disturb defects. None Of the disturb-causing defects could be detected with IDDQ.
VII. Conclusions In this paper, we studied the power supply current detectability (both transient and quiescent) of various SRAM defects. These defects include shorts, opens, and disturbs. It was shown that when combined together, IDDQ and iDDT form a strong test vehicle for reducing the number of escaping defects. We also presented data on the effects of large circuit parameters such as large leakages and capacitance loading. This data can be used to estimate the size of partitions for current testing.
References L. Peters, “A Better Method for Testing CMOS ICs” Semiconductor International, November 1991, page 36. W. Maly, “Realistic Fault Modeling for VLSI Testing,” Proc. of Design Automation Conference, 1987, pp. 173-180. J. Soden and C. Hawkins, “Test Considerations for Gate Oxide Shorts in CMOS ICs,” IEEE Design and Test, Aug. 1986, pp. 56-63. T. Storey, W. Maly, J. Andrews, and M. Miske, “Stuck Fault and Current Testing Comparison Using CMOS Chip Test,” Proc. of International Test Conference, 1991,pp. 311-318. W. Maly, “Current Testing,” Proc. of International Test Conference, 1990,page 257. J. Soden, R. Fritzemeier, and C. Hawkins, “Zero Defects or Zero Stuck-At Faults - CMOS IC Process Improvement with IDDQ,~’Proc. of International Test Conference, 1990,pp. 255-256. S . Bollinger and S . Midkiff, “On Test Generation for IDDQ Testing of Bridging Faults in CMOS Circuits,” Proc. of International Test Conference, 1991, pp.
372
Authorized licensed use limited to: University of Central Florida. Downloaded on October 25, 2008 at 23:37 from IEEE Xplore. Restrictions apply.
International Conference on CAD, 1990,pp. 280-283. [231 P. Nigh and W. M y , “Layout-Driven Test Generation,” Proc of International Conference on CAD,1989,pp. 154-157. [24] R. Fritzemeier, J. Soden, R. Treece, and C. Hawkins, “Increased Stuck-at Fault Coverage with Reduced IDDQ Test Sets,” Proc. of International Test Conference, 1990. [25] T. Chakraborty, S . Bhawmik, R. Bencivenga and C. Lin, “Enhanced Controllability For IDDQ Test Sets Using Partial Scan,” Proc. of 28th ACMIIEEE Design Automation Conference, pp. 278-281. [26] L. Homing, J. Soden, R. Fritzemeier, and C. Hawkins, “Measurement of Quiescent Power Supply Current for CMOS ICs in Production Testing,” Proc. of International Test Conference, 1987,pp. 300-309. [27] R. Actken, “Fault Location with Current Monitoring,” Proc. of International Test Conference, 1991, pp. 623-632. [28] R. Meershoek, B. Verelst, R. Mclnerey, and L. Thijssen, “Functional and IDDQ Testing on a Static RAM,”Proc. of International Test Conference, 1990, pp. 929-937 [29] C. Kuo, T. Toms, B. Neel, J. Jelemensky, E. Carter, and P. Smith, “Soft-Defect Detection (SDD) Technique for a High-Reliability CMOS SRAM,” IEEE Journal of Solid-State Circuits, Vol. 25, No. 1, Feb. 1990, pp. 61-66. [30] K. Baker, “QTAG: A Standard for Test Fixture based IDDo/lssaMonitors,” Proc of ITC 94 pp. 194-202. [31] S . Su and R. Makki, “Testing Random Access Memories By Monitoring Dynamic Power Supply Current,” Journal of Electronic Testing, Vol. 3, Aug. 1992, pp. 265-278. [32] S . Su, R. Makki, and T. Nagle, “Transient Power Supply Current Monitoring - A New Test Method for CMOS VLSI Circuits,” Journal of Electronic Testing, Vol. 6, February 1995, pp. 23 - 43. [33] S . Su and R. Makki, “A Testable Static RAM Structure for Efficient Coverage of Pattem Sensitive Faults,” Proc. IEEE VLSI Test Symposium, Apr. 1992, Session 12.4. [34] W. Maly and M. Patyra, “Design of ICs Applying Built-In Current Testing,” Journal of Electronic Testing, Vol. 3, December 1992,pp. 397-406.
373
Authorized licensed use limited to: University of Central Florida. Downloaded on October 25, 2008 at 23:37 from IEEE Xplore. Restrictions apply.