Board-Level Test Technologies Extending IEEE Std. 1149.4 Analog Boundary Modules to Enhance Mixed-Signal Test Uroš Kač and Franc Novak Florence Azaïs, Pascal Nouet, and Michel Renovell Jozef Stefan Institute LIRMM with support from component manufacturers as well as EDA tool and test equipment providers. Its mixed-signal twin, on the other hand, has had a much harder time finding its way into real-life applications. Although research has focused on IEEE 1149.4,5,6 and technology demonstrator chips for evaluating its proposed features exist,7,8 Dot 4 lacks support in standard catalog devices, making it difficult for designers to include a standardized mixed-signal test infrastructure in their systems. The absence of commercial Dot-4-compliant devices could be due to the test infrastructure’s complexity and its potential impact on proven analog and mixed-signal designs. The benefits of including overhead logic in the circuit must justify the effort required to do so. For example, applications aided IEEE 1149.1’s acceptance by demonstrating its value not only in the design testability domain but also in areas such as design validation, debugging, and in-system configuration of programmable devices.9 The work of Sunter and colleagues on a general-purpose Dot-4-compliant IC attempts to improve the standard’s status.9,10 As designers demonstrate the benefits of practical mixed-signal designs and present interesting applications based on the Dot 4 infrastructure, interest in the standard will increase. Several researchers have already reported innovative Dot-4-based measurement procedures.9,11,12 We describe a Dot 4 test chip with extensions to the original 1149.4 test infrastructure that expands possible practical applications of the standard. Editor’s note: Will it or won’t it? The 1999 IEEE 1149.4 Standard for a Mixed-Signal Test Bus is on the cusp of industrial acceptance, but it’s not clear whether industry will pick it up. This study, by two leading European research institutes, delves into the details of hardware implementation and, in so doing, contributes to the growing literature on this topic. —R.G. (Ben) Bennetts, Bennetts Associates INCREASINGLY DENSE and complex electronic designs have made established in-circuit test (ICT) techniques more costly and difficult to implement. Several electronic systems manufacturers, such as Philips and IBM, have proposed an innovative boundary scan method to improve design controllability and observability. The method is based on ICT techniques, but it substitutes physical nails with an on-chip shift register placed around the IC core boundary. This boundary scan register captures IC input signals and applies test vectors on IC outputs through its parallel I/Os. Test data is shifted in and out through a simple serial interface, eliminating the need for direct physical access.1 In 1990, as part of the Joint Test Action Group’s (JTAG) initiative, the IEEE adopted the approach as IEEE 1149.1.2 Because IEEE 1149.1 addressed only digital circuits, designers soon began developing equivalent test structures for mixed-signal circuits and corresponding measurement methodologies.3 Standardization efforts on an IEEE-1149.1compatible test bus that would improve mixed-signal design testability at both the device and the assembly level led to the IEEE 1149.4 (known as Dot 4) standard in 1999.4 The IEEE 1149.1 boundary scan standard has become a widely accepted design for digital circuits, 32 0740-7475/03/$17.00 © 2003 IEEE Copublished by the IEEE CS and the IEEE CASS IEEE Design & Test of Computers Implementing the Dot 4 test chip Common reference voltage VH Pin5 Core5 Pin4 Core4 Pin3 VTH ABM5 ABM6 ABM4 ABM7 ABM3 ABM8 Pin6 Core6 Pin7 Core7 Pin8 Core8 AB1 Core3 VG VL Pin2 ABM2 Core2 Pin1 ABM1 Core1 Pin9 ABM9 Core9 AB2 We approached the design of the Dot 4 test chip in two stages. Our preliminary chip, shown in Figure 1, adheres to the standard IEEE 1149.4 architecture. Thus, we can assess the actual characteristics of complex mixed-signal cells—analog boundary modules (ABMs) and a test bus interface circuit (TBIC)—and identify possible design inefficiencies and necessary modifications. To simplify design debugging, we implemented the test access port (TAP) controller, instruction and bypass registers, and decoder off chip with a programmable logic device (PLD). The design follows schematic representations of the boundary scan register modules proposed in the IEEE 1149.4 standard. We implemented three main cells: the ABM, the TBIC, and the test control circuitry, each comprising several submodules. Because we wanted users to be able to test the Dot 4 infrastructure in their own designs, we didn’t include an analog functional core. Although IEEE 1149.4 lets designers implement conceptual switches differently, transmission gates are most appropriate for CMOS-based devices. Figure 2 shows an ABM cell’s analog-switching architecture. We implemented the preliminary chip in 0.8-micron Austria Microsystems (AMS) CYE technology. The relatively high internal resistance exhibited by standard AMS library switch cells required that we modify the design to obtain transmission gates suitable for inclusion in the ABM/TBIC analog-switching architecture. Similarly, because standard cell comparators would occupy an unacceptably large silicon area, we designed a custom comparator, which satisfies our requirements both in terms of silicon area and electrical characteristics. Table 1 compares the library and modified switched cell. Table 2 compares the standard library and custom comparators. AT1 AT2 TBIC Dot 4 test chip TDI Mode1 TDO Update_DR Clock_DR Shift_DR Mode2 Common test control signals TAP controller Complex programmable logic device AB ABM AT TAP TBIC TDI TDO Analog bus Analog boundary module Analog test pin Test access port Test bus interface circuit Test data input Test data output Figure 1. Preliminary test chip block diagram. We use this chip to assess complex mixedsignal cell characteristics and identify design deficiencies and modifications. Table 1. Comparison of area and resistance between library and modified switch cell. Cell Surface area (µm2) Dimensions (µm) RON at VDD (Ω) TG2B (library) 16.2 × 34.5 559 1,620 TG_inv (modified, integrated inverter) 26.8 × 39.7 750 750 Table 2. Comparison between library and custom comparator implementations. Offset Voltage Vout min Vout max Vth min Vth max Surface (µV) gain AV (dB) (mV) (V) (V) (V) area (µm2) Comp01B (library) 27 87 3 5 0.1 4.4 = 23,000 Comparator (full custom) 92 61 17 5 0.1 4.6 = 2,800 Cell March–April 2003 33 Board-Level Test Technologies Reference voltages VH VTH VL modified the ABM cell in three key ways: VG ■ SH SL Switching architecture SG SD Pin Core SB1 SB2 AB2 AB1 Control-decoding logic Mode2 Mode1 ■ ■ Update_DR Clk D D Shift_DR G D Clk D D D Clk B1 D D Clk B2 D D Update register To TDO We excluded the core disconnect switch from the ABM switching architecture and implemented it as a separate cell consisting of a lowresistance (RON < 50 Ω) CMOS transmission gate, where RON is the transmission gate resistance during the ON state. Users can take advantage of this low-resistance switch or implement the core disconnect facility in their own analog designs. We added two switches to provide ABM analog test bus multiplexing. We added switches to connect the ABM comparator’s inverting input to either VTH or one of the VH, VL, or VG common reference voltages. Applying the extended ABM functionality Figure 4a shows how digital boundary modules isolate the core from external cir0 From TDI Clk Clk Clk Clk cuitry during internal test (Intest). During Clock_DR digital Intest, digital boundary modules Analog boundary module isolate the core from the I/O pins while the tester applies test vectors and moniFigure 2. Schematic representation of an analog boundary module (ABM). tors responses. On the other hand, accordWe divided the ABM into four submodules to simplify future modifications ing to IEEE 1149.4, the analog INTEST of the design. instruction keeps external circuitry connected to the analog core. This requires Synthesizing the control logic and registers was rel- controlling the external onboard circuitry to provide atively straightforward. Figure 3 shows the test chip lay- appropriate operating conditions to the core or to ensure out. Altogether, we laid nine ABM cells and one TBIC that the inputs are quiescent. To circumvent this practical limitation in the origon an active area of 1,980 × 1,980 square microns, featuring 39 pads connected to five control signals (Mode1, inal ABM structure, we introduce two additional Mode2, Update_DR, Shift_DR, and Clock_DR); two digi- switches, SB1_INT (internally switch to bus 1, AB1) tal test signals (TDI and TDO); two digital power supply and SB2_INT, which bypass the core disconnect lines (VDD and GND); two analog power supply lines switch (SD), as Figure 4c illustrates. This lets design(VDDA and GNDA); two analog test signals (AT1 and AT2); ers open SD during the INTEST instruction, eliminatand 18 pairs (pin and core) of ABM functional signals. ing the need to control external circuitry and We directly synthesized the TAP controller, bypass and simplifying test development. The proposed ABM modification does not replace instruction registers, and instruction decoding logic from RTL VHDL source code and programmed them into a the original INTEST instruction but rather adds an Xilinx XC95108 complex PLD for a total of 38 macro cells. instruction, letting the designer perform Intest with the The second version of the Dot 4 test chip incorpo- core disconnected from the surrounding circuitry. For rates both analog and digital components and includes extended Intest procedures on a multiple input analog extensions that let us implement alternative test and core, this approach is limited: The analog stimulus genmeasurement procedures. To realize this version, we erator can drive only a single input through the analog 1 34 Data CTRL Bus1 Bus2 Control register IEEE Design & Test of Computers VL Pin3 Pin7 Core1 Core7 Pin2 Pin8 Core1 Core8 Pin1 AT1 Core0 AT2 Pin0 VDDA VDDA VSS VSS Shift_DR D Shift_DR Clock_DR VSS Figure 3. Preliminary test chip layout. Nine ABMs and one TBIC cell are laid out starting from the bottom left. To simplify debugging, we did not include the TAP controller on the chip. VH VL VG G SH Core SL SG SD Pin D Pin ON Core 1 Clk From TDI Mode2 VDD Mode1 VSS Scan_in V VDD VTH 0 1 G Pin6 Core6 Mode Pin OFF VG Core3 To TDO Pin VTH Pin4 Core4 Pin5 Core VH 5 VH Scan_out test bus. We therefore include switch SG_INT, which connects the remaining core inputs to the known reference voltage, VG, as Figure 4c shows. The original and modified ABM architectures suit different types of test strategies. The modified ABM structure allows system-wide functional reconfiguration, which might prove useful in analog functional tests. As Kac et al. have demonstrated, designers can reconfigure select analog functional blocks into a selftesting structure by establishing connections via the Dot 4 analog test bus, and thus perform efficient go/nogo functional tests.13 Consider, for example, a device with an active resistor-capacitor (RC) filter core, as depicted in Figure 5 (next page). The core input and output are connected to ABMs. By connecting the filter to an external circuit (an equalizer) via the Dot 4 analog test bus, we can perform an oscillation-based test. We can then measure the resulting frequency onboard with a simple digital counter and compare it to a preloaded reference value. An experiment verified this idea’s feasibility. Figure 6 shows the resulting oscillation, and Table 3 gives measurement results, where Rb determines the equalizer voltage gain and fOSC denotes the resulting oscillation frequency. Update_DR Clk External load Clock_DR (a) (b) VTH VH SH VL SL SB1 SB2 AB1 AB2 VG SG Pin Pin OFF SD SG _INT Core External load (c) SB1 SB2 S1_INT S2_INT AB1 AB2 Figure 4. Isolating the core during Intest procedures: digital boundary module (a), ABM switching architecture (b), and modified ABM switching architecture (c). March–April 2003 35 Board-Level Test Technologies IC ABM1 Analog core (RC filter) ABM 2 AB2 External circuitry AB1 AT1 TBIC Equalizer System test bus AT2 Figure 5. Active resistor-capacitor (RC) filter functional test. We connect the filter to an equalizer using the Dot 4 analog test bus, then perform the oscillation-based test. VH IC1 Voltage comparator logic signal (Cmp) VTH IC2 Vshort ABM11 RSHORT<<RON+RPAD SH CMOS switch RON Out1 VTHvariations ABM21 Cmp21 In1 1 Pad RPad VshortH CPad PAD VshortL Pad ABM12 RON Pad Out2 In2 Cmp22 SL (a) ABM22 (b) VL Vin 0 VL VTHmin VTH VTHmax VH Figure 6. IEEE 1149.4 interconnect test (Extest): A bridging fault produced an intermediate-level voltage on analog input pins (a). Variations in ABM comparator threshold levels can lead to a fault passing undetected (b). Table 3. Measurement results for direct and analog test bus connections. Connection type Rb (kΩ) fOSC (Hz) Direct 241 1323 Via analog test bus 297 1323 In the original IEEE 1149.4 ABM structure, the control-decoding logic derives the control signals for each switch from the update register contents and the global mode signals, both of which the TAP 36 instruction decoder supplies (Figure 2). The extended ABM switching architecture requires additional control signals, which can be provided in different ways. Because switches (SB1, SB2, and SG) and (SB1_INT, SB2_INT, and SG_INT) represent two complementary groups, they can use the same update register codes for the control, provided the instruction decoder supplies an additional global control signal (Mode3) during the extended INTEST instruction. Table 4 shows how we modified the equations for this instruction. IEEE Design & Test of Computers The hardware overhead is therefore relatively small, encompassing the additional ABM switches, the modified ABM control-decoding logic, and the additional global control line. Because switches SB1_INT, SB2_INT, and SG_INT introduce additional parasitic capacitances to the signal path, minimizing their impact on the analog design functionality requires careful modeling. Our second modification to the ABM cell lets us compare analog input signals to multiple voltage levels. We add switches CG, CL, CH, and CTH, where C is compare, to the ABM switching structure, as Figure 7a illustrates. These switches connect the comparator inverting input to either VTH (mandatory) or one of the VH, VL, or VG common reference voltages (our extension). This feature augments the diagnostic capability of interconnect test (Extest). Assume the tester encounters a bridging fault during Extest, as Figure 6a illustrates. In the preliminary chip implementation (see Figure 2), we implemented SH and SL with identical CMOS transmission gates. These gates exhibit substantial ON-resistance and are connected in series with reference voltage sources. As IEEE 1149.4 notes, the voltage at the combined (shorted) net will be at some value between VH and VL. Designers should choose a compare voltage, VTH, that is clearly different from the possible input voltage levels. On the other hand, IEEE 1149.4 limits the value of VTH value to the range, (VH + VL)/2 ± (VH – VL)/4. Suppose the resulting voltage of the bridging fault depicted in Figure 6a is near the threshold voltage VTH of IC2. As Table 4. Control equations for the extended INTEST instruction, where B, C, and D refer to update register cells (shown in Figure 2), and M refers to mode. Switch SG Original equation – SG = CDM1M2 SB1 SB1 = B1M1 SB2 SB2 = B2M2 SG_INT NA SB1_INT NA SB2_INT NA Modified equation – –– SG = CDM1M2M3 –– –– SB1 = B1M1M2 M3 –– –– SB2 = B2M1M2 M3 – –– SG_INT = CDM1M2M3 –– SB1_INT = B1M1M2M3 –– SB2_INT = B2M1M2M3 Figure 6b shows, variations in comparator threshold levels cause the ABM21 comparator to identify the input voltage as high logic and the ABM22 comparator as low logic. Hence, the fault can pass undetected. If we shift the comparator threshold voltage to level VH2 in ABM21 and VL2 in ABM22, we can clearly detect the bridging fault, as Figure 7b illustrates. Dot 4 states that VH and VL should always be available on both input and output pins and can be pin specific—that is, they do not have to be the same on all pins. Therefore, designers can choose appropriate VH2 and VL2 values for IC2 input pins (In1, In2) such that the comparators maintain a sufficient noise margin for both fault-free input voltage levels (VH and VL). Because Extest does not require high-precision voltage levels, simple MOS voltage dividers can generate VH2 and VL2 on chip. Alternatively, external sources can supply the required multiple voltVoltage comparator logic signal (Cmp) VG VL VH VTH VH2 IC2 ABM21 ABM CG CL CH CTH SG SL SH Vcmp Pad Cmp SB1 (a) Cmp22 ABM22 Vcmpcontrol VTH VH2 VH1 1 Pad In2 AB1 AB2 VIn1 0 VL1 Core SB2 Vshort Cmp21 Vshort SD Vin Cmp21 In1 1 VL2 Cmp22 Vshort VIn2 0 VL1 VL2 VTH VH1 (b) Figure 7. Controlling VTH during interconnect test (Extest): Added switches CG, CL, CH, and CTH connect the comparator inverting input to either VTH or VH, VL, or VG (a); shifting the comparator threshold voltage to level VH2 in ABM21 and VL2 in ABM22 makes the bridging fault easily detectable (b). March–April 2003 37 Board-Level Test Technologies FUTURE WORK will focus on functional reconfiguration strategies exploiting the proposed core disconnect feature. In particular, we shall explore possible reconfigurations of different types of analog circuits into oscillation-based test structures. We also plan to work on applications of the modified ABMs that allow testers to compare analog input signals with multiple voltage levels. In addition to the enhanced interconnect test we describe, the comparator input can be connected to different sensors, thus allowing the system to monitor its environment via the Dot 4 infrastructure. ■ Table 5. Control equations for the modified EXTEST instruction. Switch SH Original equation SG SH = CDM1M2 – SL = CDM1M2 – SG = CDM1M2 CH NA CL NA CG NA CTH NA SL Modified equation –– SH = CDM1M2M3 – –– SL = CDM1M2M3 – –– SG = CDM1M2M3 CH = CDM1M2M3 – CL = CDM1M2M3 – CG = CDM1M2M3 –– CTH = M3 Acknowledgments ages through additional, common voltage reference pins. Like the modified INTEST instruction, the modified EXTEST instruction requires additional control signals, which can be similarly provided. Switches SH, SL, and SG on the input pins are always open during conventional Extest; consequently, a common signal can disable them and the modified EXTEST instruction can use their update register codes to control switches CH, CL, and CG. Again, we combine the global mode signals with the codes to produce the switch control signals, as Figure 7 shows. Table 5 shows how we changed the control equations to support the modified EXTEST instruction. Figure 8 depicts a trivial implementation of the modified ABM control-decoding logic. We performed this work as part of the bilateral French-Slovenian Proteus project, which is supported by the French Ministry of Foreign Affairs and the Slovenian Ministry of Education, Science, and Sport. References 1. R.G. Bennetts, IEEE 1149.1 Test Access Port and Boundary-Scan Std., DFT Technology Backgrounders, Feb. 2001, http://www.semiconductorfabtech.com/dft/ tutorial/bscan.PDF. 2. IEEE Std. 1149.1-2001, IEEE Standard Test Access Port and Boundary-Scan Architecture, IEEE, 2001. 3. K.P. Parker, J.E. McDermit, and S. Oresjo, “Structure and Metrology for an Analog Testability Bus,” Proc. Int’l Test Conf. (ITC 93), IEEE Press, 1993, pp. 309-317. SD SG_INT CG CL CH SG SL SH SG2_INT SB2 SB1 SG1_INT CTH Switch control signals Control-decoding logic Mode1 Mode2 Mode3 Update_DR Clk D D Clk C D Clk B1 D Update register Clk B2 D From control register Figure 8. Modified ABM control-decoding logic for extended Intest and Extest. 38 IEEE Design & Test of Computers 4. IEEE Std. 1149.4-1999, IEEE Standard for a MixedSignal Test Bus, IEEE, 1999. 5. A. Cron, “IEEE 1149.4—Almost a Standard,” Proc. Int’l Test Conf. (ITC 97), IEEE Press, 1997, pp. 174-182. 6. K.P. Parker, The Boundary-Scan Handbook: Analog and Digital, Kluwer, 1998. 7. “JTAG Analog Extension Test Chip: Target Specification for the IEEE P1149.4 Working Group,” Preliminary Review 012, Keith Lofstrom Integrated Circuits, Beaver- Franc Novak heads the Computer Systems Department at the Jozef Stefan Institute and is an associate professor on the Faculty of Electrical Engineering and Computer Science at the University of Maribor. His research interests include electronic testing and diagnosis, fault-tolerant computing, and DFT of analog circuits. Novak has a PhD in electrical engineering from the University of Ljubljana. ton, Ore., 1998; http://grouper.ieee.org/groups/1149/4/ kl1p.html. 8. K.P. Parker et al., “Design, Fabrication, and Use of Mixed-Signal IC Testability Structures,” Proc. Int’l Test Conf. (ITC 97), IEEE Press, 1997, pp. 489-498. 9. S. Sunter et al., “A General Purpose 1149.4 IC with HF Analog Test Capabilities,” Proc. Int’l Test Conf. (ITC 01), IEEE Press, 2001, pp. 38-45. 10. National Semiconductor, “Mixed-Signal Test, and the IEEE 1149.4 Standard,” http://www.national.com/ appinfo/scan/. Florence Azaïs is a researcher for the National Council for Scientific Research (CNRS) in the Microelectronics Department of LIRMM (the Laboratory of Computer Science, Robotics, and Microelectronics in Montpellier, France). Her research interests include fault modeling and mixed-signal circuit testing, especially DFT and BIST techniques. Azaïs has a PhD in electrical engineering from the University of Montpellier. 11. K. Lofstrom, “Early Capture for Boundary Scan Timing Measurements,” Proc. Int’l Test Conf. (ITC 96), IEEE Press, 1996, pp. 417-422. 12. S. Sunter and B. Nadeau-Dostie, “Complete, Contactless I/O Testing—Reaching the Boundary in Minimizing Digital IC Testing Cost,” Proc. Int’l Test Conf. (ITC 02), IEEE Press, 2002, pp. 446-455. 13. U. Kač et al., “Alternative Test Methods Using IEEE 1149.4,” Proc. Design, Automation, and Test in Europe (DATE 00), IEEE CS Press, 2000, pp. 463-467. Uroš Kač is a research assistant at the Jozef Stefan Institute in Ljubljana, Slovenia, where he is also pursuing a PhD. His research interests include high-level synthesis, mixed-signal test, and online/concurrent built-in self-test (BIST). Kač has an MS in electrical engineering from the University of Ljubljana. March–April 2003 Pascal Nouet is a researcher at LIRMM and an associated professor at the University of Montpellier’s Institute for Engineering Sciences. His research interests include analog circuit design and test, and design, test, modeling, and characterization of monolithic microelectromechanical systems. Nouet has a PhD in microelectronics from the University of Montpellier. Michel Renovell heads the Microelectronics Department at LIRMM. His research interests include fault modeling, analog testing, and FPGA testing. Renovell has a PhD in electrical testing from the University of Montpellier. Direct questions and comments about this article to Uroš Kač, Jozef Stefan Institute, Jamova 39, 1000 Ljubljana, Slovenia; uros.kac@ijs.si. 39