Diode-turn-on-time-induced failures in switching regulators
Most circuit designers are familiar with diode dynamic characteristics, such as charge storage, voltage-dependent capacitance, and reverse recovery time. Manufacturers less commonly acknowledge and specify diode forward turn-on time. This parameter describes the time required for a diode to turn on and clamp at its forward-voltage drop. Historically, this extremely short time, measured in nanoseconds, has been so small that user and vendor alike have essentially ignored it. They rarely discuss and almost never specify it. Recently, switching-regulator clock rate and transition time have become faster, making diode-turn-on time a critical issue. The industry has mandated increased clock rates to achieve smaller magnetics; decreased transition times somewhat aid overall efficiency but are principally needed to minimize IC heat rise. At clock speeds beyond about 1 MHz, transition-time losses are the primary sources of die heating.
A potential difficulty due to diode-turn-on time is that the resultant transitory overshoot voltage across the diode, even when restricted to nanoseconds, can induce overvoltage stress, causing switching-regulator-IC failure. As such, careful testing is required to qualify a given diode for a particular application to ensure reliability. This testing, which assumes low-loss surrounding components and layout in the final application, measures turn-on overshoot voltage due to diode parasitics only. Improper associated component selection and layout will contribute additional overstress terms.
Diode turn-on time
Figure 1 shows typical step-up and step-down voltage converters. In both cases, the assumption is that the diode clamps switch pin-voltage excursions to safe limits. In the step-up case, the switch pin’s maximum allowable forward voltage defines this limit. The switch pin’s maximum allowable reverse voltage sets the step-down case limit.
Figure 2 indicates that the diode requires a finite length of time to clamp at its forward voltage. This forward-turn-on time permits transient excursions above the nominal diode-clamp voltage, potentially exceeding the IC’s breakdown limit. You typically measure the turn-on time in nanoseconds, making observation difficult. A further complication is that the turn-on overshoot occurs at the amplitude extreme of a pulse waveform, precluding high-resolution amplitude measurement. You must consider these factors when designing a diode-turn-on-test method.
Figure 3 shows a conceptual method for testing diode-turn-on time. Here, the test is performed at 1A, although you could use other currents. A pulse steps 1A into the diode under test via the 5Ω resistor. You measure turn-on-time-voltage excursion directly at the diode under test. The figure is deceptively simple. In particular, the current step must have an exceptionally fast, high-fidelity transition, and faithful turn-on time determination requires substantial measurement bandwidth.
Figure 4 offers a more detailed measurement scheme. The design requires a less-than-1-nsec-rise-time pulse generator; a 1A, 2-nsec-rise-time amplifier; and a 1-GHz oscilloscope. These specifications represent realistic operating conditions; you may select other currents and rise times by altering appropriate parameters (see sidebar “Connections, cables, attenuators, probes, and picoseconds”).
The pulse amplifier necessitates careful attention to circuit configuration and layout. As Figure 5 shows, the amplifier includes a paralleled, Darlington-driven RF-transistor-output stage. The collector-voltage adjustment, or rise-time trim, peaks Q4 to Q6 FT; an input-RC network optimizes output-pulse purity by slowing the input-pulse rise time to within the amplifier passband. Paralleling allows Q4 to Q6 to operate at favorable individual currents, maintaining bandwidth. When you optimize the mildly interactive edge-purity and rise-time trims, Figure 6 indicates, the amplifier produces a transcendently clean 2-nsec rise-time output pulse devoid of ringing, alien components, or post-transition excursions. Such performance makes diode-turn-on-time testing practical (see sidebar “Verifying rise-time-measurement integrity”).
Figure 7 depicts the complete diode-forward-turn-on-time-measurement arrangement. The pulse amplifier, driven by a subnanosecond pulse generator, drives the diode under test. A Z0 probe monitors the measurement point and feeds a 1-GHz oscilloscope.
The measurement-test fixture, properly equipped and constructed, permits diode-turn-on-time testing with excellent time and amplitude resolution. Figure 8, Figure 9, Figure 10, Figure 11, and Figure 12 show results for five diodes from various manufacturers. Figure 8 (Diode 1) overshoots steady-state forward voltage for 3.6 nsec, peaking at 200 mV, offering the best performance of the five. Figure 9 through Figure 12 show increasing turn-on amplitudes and times. In the worst cases, turn-on amplitudes exceed nominal clamp voltage by more than 1V, and turn-on times extend for tens of nanoseconds. Figure 12 culminates this unfortunate parade with huge time and amplitude errors. Such errant excursions can and will cause IC-regulator breakdown and failure. The lesson here is clear: You must characterize and measure diode turn-on time in any given application to ensure reliability.
Connections, cables, adapters, attenuators, probes, and picoseconds
You must consider rise-time signal paths of less than 1 nsec as transmission lines. Connections, cables, adapters, attenuators, and probes represent discontinuities in this transmission line, deleteriously affecting its ability to faithfully transmit the desired signal. The degree of signal corruption that a given element contributes varies with its deviation from the transmission line's nominal impedance. The practical result of such aberrations is degradation of pulse rise time, fidelity, or both. Accordingly, you should minimize the introduction of elements or connections to the signal path and use high-grade components for connections and elements.
Any form of connector, cable, attenuator, or probe must be fully specified for high-frequency use. Familiar BNC hardware becomes lossy at rise times much faster than 350 psec. SMA components are preferable for the rise times this article describes. Additionally, to minimize inductance, cable-induced mismatch, and distortion, connect the pulse-amplifier output directly to the diode under test without using cable. Avoid mixing signal-path hardware types via adapters. Adapters introduce significant parasitics, resulting in reflections, rise-time degradation, resonances, and other degrading behavior. Similarly, make oscilloscope connections directly to the instrument's 50Ω inputs, avoiding probes. If you must use probes, introduction to the signal path mandates attention to their connection mechanism and high-frequency compensation. Passive low-impedance types, commercially available in 500Ω and 5-kΩ impedances, have input capacitance of less than 1 pF (see sidebar "About low-impedance probes"). You must carefully frequency-compensate any such probe before use, or measurement may be misrepresented. Inserting the probe into the signal path necessitates some form of signal pick-off, which nominally does not influence signal transmission. In practice, some amount of disturbance must be tolerated and its effect on measurement results evaluated. High-quality signal pick-offs always specify insertion loss, corruption factors, and probe-output scale factor.
Be vigilant when designing and maintaining a signal path. Skepticism, tempered by enlightenment, is a useful tool when constructing a signal path, and no amount of hope is as effective as preparation and directed experimentation.
|How much bandwidth is enough?|
Accurate wideband-oscilloscope measurements require bandwidth. Just how much do they need? A classic guideline is that end-to-end measurement-system rise time is equal to the root-sum-square of the system's components' rise times. The simplest case is two components: a signal source and an oscilloscope. Figure A's plot of rise time versus error is illuminating. The figure plots signal-to-oscilloscope rise-time ratio versus observed rise time. Rise time is bandwidth restated in the time domain, where rise time (nsec)=350/bandwidth in megahertz.
The curve shows that measurement accuracy inside about 5% requires an oscilloscope three to four times faster than the input-signal rise time. Therefore, trying to measure a 1-nsec-rise-time pulse with a 350-MHz oscilloscope with a rise time of 1 nsec leads to erroneous conclusions. The curve indicates a monstrous 41% error. Note that this curve does not include the effects of passive probes or cables connecting the signal to the oscilloscope. Probes do not necessarily follow root-sum-square law, and you must carefully choose and apply them for a given measurement. Table A gives 10 cardinal points of rise-time/bandwidth equivalency between 1 MHz and 5 GHz.
Figure B through Figure I illustrate pertinent effects of these considerations by viewing the main article's diode-turn-on-time measurement at various bandwidths. displays a typical diode turn-on in a 2.5-GHz sampled bandpass, showing 500-mV turn-on amplitude. Figure C's 1-GHz bandwidth measurement has nearly identical characteristics, indicating adequate oscilloscope bandwidth. The dramatic error in observed turn-on overshoot amplitude as bandwidth decreases in succeeding figures is readily apparent and should not be lost to the experimenter.Figure B
Less-than-1-nsec-rise-time pulse generators for the rich and poor
The pulse amplifier requires a less-than-1-nsec-input-rise-time pulse to cleanly switch current to the diode under test. Most general-purpose pulse generators have rise times of 2.5 to 10 nsec. Instrument rise times of less than 2.5 nsec are relatively rare, with only a select few types getting down to 1 nsec. The ranks of less-than-1-nsec-rise-time generators are even thinner, and costs are excessive. Subnanosecond-rise-time generation, particularly if you want relatively large swings of 5 to 10V, employs arcane technologies and exotic construction techniques. Available instruments in this class work well but can easily cost $10,000, with prices rising toward $30,000 depending on features. For benchwork, or even production testing, there are substantially less expensive approaches.
The secondary market offers less-than-1-nsec-rise-time pulse generators at attractive cost. The Hewlett-Packard HP-8082A transitions in less than 1 nsec, has a full complement of controls, and costs about $500. The Tektronix type 111 has edge times of 500 psec, with fully variable repetition rate and external-trigger capabilities. External charge-line length sets the pulse width. The device usually costs about $25. The HP-215A, long out of manufacture, has 800-psec edge times and is a clear bargain, typically costing less than $50. This instrument also has a versatile trigger output, permitting continuous trigger-time phase adjustment from before to after the main output. External-trigger impedance, polarity, and sensitivity are also variable. The output, which a stepped attenuator controls, puts a clean ±10V pulse into 50Ω in 800 psec.
A potential problem with older instruments is availability. As such, Figure A shows a circuit for producing subnanosecond-rise-time pulses. Rise time is 400 psec, with adjustable pulse amplitude. Output pulse occurrence is settable from before to after a trigger output. This circuit uses an avalanche pulse generator to create extremely fast rise-time pulses.
Q1 and Q2 form a current source that charges the 1000-pF capacitor. When the LTC1799 clock is high (Trace A, Figure B), both Q3 and Q4 are on. The current source is off, and Q2's collector (Trace B) is at ground. C1's latch input prevents it from responding, and its output remains high. When the clock goes low, C1's latch input is disabled and its output drops low. The Q3 and Q4 collectors lift, and Q2 comes on, delivering constant current to the 1000-pF capacitor (Trace B). The resulting linear ramp is applied to C1's and C2's positive inputs. C2, biased from a potential derived from the 5V supply, goes high 30 nsec after the ramp begins, providing the "trigger output" (Trace C) via its output network. C1 goes high when the ramp crosses the potentiometer-programmed delay at its negative input, in this case, about 170 nsec. C1 goes high, triggering the avalanche-based output pulse (Trace D). This arrangement permits the delay programming control to vary output-pulse occurrence from 30 nsec before to 300 nsec after the trigger output. Figure C shows the output pulse (Trace D) occurring 25 nsec before the trigger output. All other waveforms are identical to those in Figure B.
When you apply C1's output pulse to Q5's base, it avalanches. The result is a quickly rising pulse across Q5's emitter-termination resistor. The collector capacitors and the charge line discharge, Q5's collector voltage falls, and breakdown ceases. The collector capacitors and the charge line then recharge. At C1's next pulse, this action repeats. The capacitors supply initial pulse response, with the charge lines' prolonged discharge contributing the pulse body. The 40-in. charge line forms an output pulse width about 12 nsec in duration.
Avalanche operation requires high voltage bias. The LT1533 low-noise switching regulator and associated components supply this high voltage. The LT1533 is a push-pull output switching regulator with controllable transition times. Output harmonic content, or "noise," is notably reduced with slower switch transitions. Resistors at the RCSL and RVSL pins, respectively, control switch current and voltage-transition times. In all other respects, the circuit behaves as a classical push-pull, step-up converter.
Circuit optimization begins by setting the output-amplitude vernier to maximum and grounding Q4's collector. Next, set the avalanche-voltage adjust so free-running pulses appear only at Q5's emitter, noting the bias test points' voltage. Readjust the avalanche-voltage adjust 5V below this voltage and unground Q4's collector. Set the 30-nsec trim so the trigger output goes low 30 nsec after the clock goes low. Adjust the delay-programming control to maximum and set the 300-nsec calibration, so C1 goes high 300 nsec after the clock goes low. Slight interaction between the 30- and 300-nsec trims may require repeating their adjustments until both points are calibrated.
Q5 requires selection for optimal avalanche behavior. The manufacturer does not guarantee such behavior, although characteristic of the device specified. A sample of 30 2N2501s, over a 17-year date-code span, yielded approximately 90%. All good devices switched in less than 475 psec, with some switching in less than 300 psec. In practice, you should select Q5 for in-circuit rise time less than 400 psec. Then, optimize the output-pulse shape by adjusting Q5's collector damping trims, including edge time/peaking and ringing.
The trims are somewhat interactive but not unduly so, and optimal adjustment converges nicely. The pulse edge is carefully adjusted so that the design attains maximum transition speed with minimal sacrifice of pulse purity. Figure D through Figure F detail the optimization procedure. In Figure D, the trims are set for significant effect, resulting in a reasonably clean pulse but sacrificing rise time. Figure E represents the opposite extreme. Minimal trim effect accentuates rise time but promotes post-transition ring. Figure F's compromise trimming is more desirable. This approach only slightly reduces edge rate but significantly slows post-transition ring, resulting in a 400-psec rise time with high pulse purity.
About low-impedance probes: when to roll your own and when to pay the money
Z0 (low-impedance) probes provide the most faithful high-speed-probing mechanism available for low source impedances. Their less-than-1-pf input capacitance and near ideal transmission characteristic make them the first choice for high-bandwidth oscilloscope measurement. Their deceptively simple operation invites do-it-yourself construction, but numerous subtleties mandate difficulty for prospective constructors. Arcane parasitic effects introduce errors as speed increases beyond about 100 MHz with a rise time of 3.5 nsec. The selection and integration of probe materials and the probes' physical incarnation require extreme care to obtain high fidelity at high speed. Additionally, the probe must include some form of adjustment to compensate for small, residual parasitics. Finally, the design must maintain true coaxiality when fixturing the probe at the measurement point, implying a high-grade, readily disconnectable, coaxial connection capability.
Figure A shows that a Z0 probe is basically a voltage-divided-input 50Ω transmission line. If R1 equals 450Ω, 10× attenuation and 500Ω input resistance result. R1 with a value of 4950Ω causes a 100× attenuation with 5-kΩ input resistance. The 50Ω line theoretically constitutes a distortionless transmission environment. The apparent simplicity seemingly permits do-it-yourself construction, but this section's remaining figures demonstrate a need for caution.
Figure B establishes a fidelity reference by measuring a clean 700-psec-rise-time pulse using a 50Ω line terminated via a coaxial attenuator, with no probe employed. The waveform is singularly clean and crisp with minimal edge and post-transition aberrations. Figure C depicts the same pulse with a commercially produced 10× Z0 probe in use. The probe is faithful, and there is barely discernible error in the presentation. Figure D and Figure E, taken with two separately constructed do-it-yourself Z0 probes, show errors. In Figure D, Probe 1 introduces pulse front-corner rounding; probe 2 in Figure E causes pronounced corner peaking. In each case, some combination of resistor/cable parasitics and incomplete coaxiality are likely responsible for the errors. In general, do-it-yourself Z0 probes cause these types of errors beyond about 100 MHz. At higher speeds, if waveform fidelity is critical, it's best to pay the money.
Verifying rise-time-measurement integrity
Any measurement requires the experimenter to ensure measurement confidence. Some form of calibration check is always in order. High-speed time-domain measurement is particularly prone to error, and various techniques can promote measurement integrity.
Figure A's battery-powered 200-MHz crystal oscillator produces 5-nsec markers, useful for verifying oscilloscope-time-base accuracy. A 1.5V AA cell supplies the LTC3400 boost regulator, which produces 5V to run the oscillator. The device delivers the oscillator output to the 50Ω load via a peaked attenuation network, which provides well-defined 5-nsec markers (Figure B) and prevents overdriving low-level sampling-oscilloscope inputs.
Once you confirm time-base accuracy, it is necessary to check rise time. This measurement should include the lumped signal-path rise time, including attenuators, connections, cables, probes, and oscilloscope. Such end-to-end rise-time checking is an effective way to promote meaningful results. A guideline for ensuring accuracy is to have a four-times-faster measurement-path rise time than the rise time of interest. Thus, the 400-psec-rise-time measurement in Figure F from the sidebar, "Less-than-1-nsec-rise-time pulse generators for the rich and poor" requires a verified 100-psec-measurement-path rise time to support it. Verifying the 100-psec-measurement-path rise time, in turn, necessitates a 25-psec-rise-time test step. Table A lists some very fast edge generators for rise-time checking.
The Hewlett-Packard 1105A/1106A, specified at 20-psec rise time, verified the measurement signal path in Figure C from the sidebar, "How much bandwidth is enough?" Figure D indicates a 140-psec rise time, promoting measurement confidence.
Another way to do it
An elegantly simple alternative method for generating the fast-rise 1A pulse is available. The Tektronix type 109 mercury-wetted reed-relay-based pulse generator puts a 50V pulse into 50Ω (1A) in 250 psec. An externally connected charge line with an approximate scale factor of 2 nsec/ft sets the pulse width. Figure A, a simplified schematic, shows type 109 operation. When the relay contacts close, the charge line discharges via the 50Ω-diode path. The pulse extends until the line depletes; depletion time depends on line length. The relay structure is arranged to assume wideband, 50Ω characteristics. Figure B shows the result. The 109 drives the monitoring 1-GHz oscilloscope to its 350-psec-rise-time limit with a 50V high-fidelity pulse.
Operating restrictions include finite relay life of approximately 200 hours; obtaining the instrument, which has been out of production for more than 20 years; difficulty in observing its low-frequency output on some oscilloscopes; and test-fixture-layout sensitivity due to the 250-psec rise time. Additionally, the faster rise time may not approximate actual circuit operating conditions as closely as the main article's 2-nsec circuit.