|Home | Articles | Forum | Glossary | Books|
As the self-appointed Czar of Band Gaps, I am impelled to continue this book with words of wisdom on voltage references, regulators, and start-up circuits. I also provide warnings against assumptions about worst-case conditions.
Voltage references and regulators have internal features that make them relatively immune to problems. But, as with other designs, if you ignore the details, you'll be headed for Trouble. Some designs incorporating these parts, such as switching power supplies, are not for the novice.
Many voltage references are based on band-gap circuitry, but some of the best references are based on buried zener diodes. If your power supply's output is in the 8-12V range, or higher, zener-diode references such as LM329, LM399, or LM369 can provide high stability, low noise, and a low temperature coefficient. If your power supply is a lower voltage (in the range from 8 down to 1.1 V) you can find band-gap references that put out a stable voltage anywhere from 0.2 to 5 V with creditable efficiency and economy. These band-gap references feature as low a temperature coefficient as you'd probably ever be willing to pay for--as good as 20 or 10 ppm / C. (They also feature enough noise so that a little filtering can make a big improvement.) A good buried-zener-diode voltage reference is inherently more stable over the long term than is a band-gap one-good zener designs change only 5 to 10 ppm per month. However, if you want the best stability possible, it's only fair to age, stabilize, and bum-in the references first. Also, you must screen out the ones that just keep "walking" away from their initial values by 10 to 20 ppm every week-there are always a few "sports" that are driftier than the rest. Unfortunately, there's no quick and easy test to distinguish between the drifty ones and the stable ones, except for taking measurements for many hundreds of hours.
Regulators Are Almost Foolproof
In the last 10 years, IC voltage regulators have gotten pretty user-friendly. Many people use them with no problems at all. Still, my colleagues and I get at least one call every month about a regulator working poorly. The indignant caller complains, "It's getting hot." We ask, "How big is your heat sink?" The indignant voice responds, "What do you mean, heat sink?" I credit all of you readers with enough smarts to recognize that you can't put a whole lot of power into a little regulator unless you secure it to a sufficient heat sink or heat fins. Then, there really aren't too many things that are likely to go wrong, because voltage regulators have just about every feature for protection against the world's assaults.
You'll have problems with regulators when you don't provide the required, specified output bypassing. Most negative regulators and some other types, such as low-dropout regulators, require an electrolytic bypass capacitor from the output to ground. If you insert a tantalum capacitor, you may be able to get away with a value of 1 or 2 pF; if you use an aluminum electrolytic capacitor, you can get away with 20 to 100 uF, or whatever the data sheet spells out. But in all cases, on all the parts I know, an electrolytic capacitor will work, and a film or ceramic capacitor won't work-its series resistance is just too small. Now, if you put a 1 ohm resistor in series with a 1 pF ceramic capacitor, the filtering will probably be adequate around room temperature; the loss factor is then similar to a tantalum capacitor. But if you take it to -40 or +100 C, the ceramic capacitor's value will shrink badly (refer to Section 4 on capacitors) and the regulator will be unhappy again. It may start oscillating, or it might just start ringing really badly.
Recently one of our senior technicians was helping a customer with some applications advice. He found that the AC output impedance of an LM317 was changing considerably as a function of the load current coming out of the output transistor. We had always assumed that the curve in the LM117 data sheet was invariant versus current load-that was a mistake. Then we found that every other monolithic regulator has the same sliding scale of output inductance. For additional notes on this phenomenon, I recommend the Erroll Dietz article (Ref. 1) which I have included because this tendency of the output inductance to be modulated by the output current may help to explain why regulators are happy in some cases but grouchy in other similar situations.
Another regulator problem can occur when you add an external transistor to increase the output current. Since this transistor adds gain at DC, it's not surprising that you have to add a big filter capacitor on the regulator's output to prevent oscillation.
Some of the applications in old National Semiconductor data sheets recommend specific values for the filter capacitor, and specific types for the boost transistors, but some of these circuits are quite old. When customers find out that 2N3234s are no longer available, they're likely to substitute a more modem transistor that has a faster response and is likely to oscillate. In this case, a customer might complain about the DC output's "bad load regulation" as the regulator is forced into and out of oscillation.
(Whoever said you don't need an oscilloscope to check out DC problems?) When these customers ask for help, I not only explain how to stop the oscillation. However, these days most engineers find it's better to use a bigger regulator (LM350 at 3 A, LM338 at 5A) because if you just add on an external transistor, you cannot protect it from overheating. Consequently the external power transistor has lost favor.
Too Much Voltage Leads to Regulator Death
You can kill any regulator with excessive voltage. So if you're driving inductive loads, or if your circuit has an inductive source, make sure to have a place for the current to go when the normal load path changes. For example, if you're using the LM350 as a simple battery charger with only a few microfarads of filter capacitor on the input, a short between the output and ground is usually disastrous: When the regulator tries to draw an increasing amount of current from the transformer and then goes into current limit, the inductance of the transformer will give you marvelous 80 V transients, which then destroy the LM350. The solution is to put 1000 pF--rather than just 1 or 10 pF--across the input.
Users get accustomed to seeing regulators with output noise of about 0.01% of the rated DC output. They get indignant when the noise doubles or triples due to 1/f or popcorn noise. The chances of finding a noisy regulator are quite small, so when some noisy ones do show up, it's a shock. Unfortunately, no high-volume manufacturer of regulators is in a position to test for those low noise levels, or to guarantee that you'll never see a noisy part. Please don't expect the manufacturer to admit the parts are bad or unreliable or worthy of being replaced. If you do depend on super-quiet ICs, or ICs with other specially selected characteristics, it's wise to keep a spare stock of selected and tested parts in a safe. Then, you can use them when some of the ones you just bought happen to be a little too noisy.
What Is Worst Case?
Once I designed a circuit to drive a 200 ohm load (a rather light load) at the far end of a 2000-ft RG174U cable. The specifications called for me to test the circuit by driving the near end of the coaxial cable with a low-impedance square wave. I called the engineer who wrote the spec and recommended that we perform the test with about a 39 ohm source impedance to avoid bad ringing and reflections along the unterminated cable. He told me that this impedance wasn't necessary; he had already checked out the worst-case conditions, with no cable and with 2000 ft of cable. I asked him if he had checked it with 250 ft of cable. "Why, no," he said. So I suggested he try that. Shortly thereafter, he called me back and agreed that the reflections with 250 ft of cable were intolerable without at least some nominal value of resistance at the source.
He had incorrectly assumed that the worst case occurred with the longest cable. It's true that the attenuation was worst with the long length of lossy RG 174U cable. But it was this attenuation that caused the ringing and reflections to appear damped out.
With the shorter 250-ft cable, a worst-case condition existed at a place he hadn't expected to find it.
So, be cautious about where you look for worst-case conditions. An op amp may exhibit its worst performance at an output voltage other than its maximum negative or positive swing--or even other than zero volts or zero output current. A regulator's worst-case operating conditions may not be at its full-rated load current. When a regulator's power source is resistive, the power dissipation may be higher at three-quarters of its rated current than at full current.
Once I worked on a regulator that ran okay at -55 C, at room temperature, and at +125 C, but not at some intermediate temperatures. That was a nasty one. Because some engineers had tested the regulators at hot and cold temperatures and saw no trouble at these extremes, I had to work very hard to convince them not to ship these parts. I had to take them by the hand and show them where the trouble was. It's like a dumb cartoon I once saw showing three men walking out of a movie-an old man, a young man, and a middle-aged man. The posters said the movie was "fun for young and old." And sure enough, the old man and the young man were smiling, but the middle-aged man was frowning. Even a dumb cartoon can be instructive if it reminds you that bad news is not only where you first expect it. It may be lurking in other places, too.
This story reminds me of a boss who asked me if my new regulator design was really short-circuit proof. I told him, yes, I had tested it with short bursts and long pulses and everything in between for days and weeks. With a wry smile, he went over to a tool cabinet and removed a really big, heavy file. He applied this file with rough, uneven scraping motions to ground and to the output of my regulator. He got showers of sparks out of the regulator, but he couldn't kill it. What a "bastard' of a test! Then he explained to me that the random, repetitive action of a file sets up patterns of current loads and thermal stresses that can kill a regulator if its short-circuit protection is marginal. There are many, many tricks you can use to show that a design really can survive every worst-case condition. Every industry has its own tests, and most of them have nothing to do with computers. . .
Switch-Mode Regulators--A Whole New Ball-Game
These simple tips aren't meant to overshadow the truly difficult areas of regulator design. You might wonder if it's possible for a smart, experienced engineer to design a switch-mode regulator that works well after only minor redesigns and goes into production without a major yield loss. My answer is: Just barely. The weasel word here is "smart." If the engineer forgets some little detail and doesn't have a contingency plan to test for it, screen it, or repair the regulators that don't work, then maybe he or she isn't very "smart." Those of us who don't design switchers all the time would have a very poor batting average at getting a design to work right off the drawing board--even if we're really good at designing other circuits. After all, a switcher is a complicated system composed of power transistors, transformers, inductors, one or more control ICs, and lots of other passive components. And, the circuit's layout is critical: The layout must guard small signals against electrostatic interference and cross-talk, and, even more importantly, must control and reject the electromagnetic strays. I mean, for a switcher to be efficient, the volts per microsecond and amperes per microsecond get really large, so it doesn't take many picofarads or nano-henries to couple a big noise into the rest of the circuit. The paths for high currents are important, and the paths for cooling air are even more critical.
So when someone asks me how to design a switcher, I ask, "How many do you plan to build?" If the answer is just a few dozen and the design is a full-featured high-power job, I encourage the engineer to buy an existing design. But if large numbers are involved, an engineer usually has the time to do the design right and spread that effort over a few thousand circuits.
An alternative to designing your own switcher is using one of the new "Simple Switchers"-the simple-but-complete switching-regulator ICs. Some of these chips--LM2575, LM2576, LM2577, LM2578--are about as foolproof-for a switcher-as you can get. The datasheets of these parts explain that. You may need a couple resistors, a few capacitors, an inductor, and a fast rectifier, and then it's done.
You'll have a cookbook circuit that really does work. And if you want to get the component selection information from a program on a floppy disk, I am told that works quite well, too, and is considered pretty ''user-friendly.'' If you only need to supply a few hundred milliamps to your circuit, you may not even need a power transistor or a heat sink. Even in the last couple years there have been advances in designs that really do work, as opposed to "paper designs" that have no chance of surviving a short on the output or of working under worst-case conditions. Although a few of these useless paper designs still pop up from time to time, thank heaven most of them have been driven out.
One of the stories that keeps rattling around the industry is about a group of engineers who decide to band together and start a new computer company. The smartest one is assigned to do the main processor board. Another smart engineer does all the interfaces. And the smart but green "kid engineer" is assigned to do the switch-mode power supply because, of course, that's the easiest part to do. (Anybody who has worked on a big switcher will probably speak up right away: The switcher is not as easy as it seems.) In the end, the power-supply design takes a lot longer than everybody expects.
One day, the young engineer opens up the compartment where the balky power supply resides, and it blows up in his face. After his co-workers take the poor fellow to the hospital, they ask around and find a consulting engineer who makes a living out of fixing exactly this kind of switcher problem. The switcher design was slightly off-course and needed the hand of an expert before it would work correctly. So remember, designing switchers is no simple task. Don't hesitate to call in an expert.
Note, if this story were not substantially true, the consulting engineer would have starved to death, long ago. I rest my case.
Regulators Suit Different Power Levels
There are several different configurations of switch-mode regulators. At low power levels, capacitively coupled switcher designs are simple, but don't provide much choice of Vout: 1.9 X Vin -0.9 X Vin, and 0.45 X Vin are almost the only choices.
Flyback regulators are the simplest and cheapest magnetically coupled regulators.
However, at power levels above 100 W, their disadvantages become objectionable, and forward or push-pull schemes are more appropriate. At the highest power levels, bridge-type designs are best. If you try to use a configuration at an inappropriate power level, you may have to struggle to get it working. Likewise, the use of current-mode regulation may help you get faster loop response, but the concept is difficult to understand, let alone execute.
Current limiting is always a problem with switchers. The choice of a sense resistor is not easy because the resistor must have low inductance. As with most aspects of switch-mode regulation, to achieve good reliability and to avoid trouble, you have to spend the time to design and test the current-limiting circuit carefully. Some newer switch-mode controller ICs have been engineered to make it reasonably easy. Older ones like the LM3524 haven't been, usually.
Similarly, a soft-start circuit is important for a large switcher, especially when the switcher strains to put out a lot of current to quickly charge up the output filter capacitors, and especially for a boost configuration, where the inductor's current might saturate and refuse to pull the output high enough. For a large supply, this current could damage transistors, wires, fuses, circuit breakers, reputations, and power companies.
A soft-start circuit forces the switcher to bring the output up to its working levels gradually, and draws only a finite amount of current from the mains as it does.
I could show you a good soft-start design, but I'll do better than that-I'll show you a bud one. The LM3524 data sheet shows (well, that's what it used to show) the circuit in FIG. 2 for a 15 V, 0.5 A step-up (Boost) switching regulator. (I should mention that this circuit and the LM3524 data sheet have been in National Semiconductor's 1978,1980, and 1982 data books; they were mistakenly left out of the 1986 data book but have been restored to the 1989 and future editions.) A "Boost" or step-up switcher needs a soft-start circuit to prevent it from saturating its transformer and from just sitting there at start-up. For this reason, C1 and D, were included to provide soft-start in FIG. 2's circuit. But FIG. 2 is still a bad circuit, if R, and D2 are not added! Let's say that the regulator is running at a low duty-cycle and the voltage on the COW pin is relatively low. Now, as the input voltage changes, the duty cycle may have to increase suddenly. But the control amplifier that drives the COMP pin not only has to pull up the series RC network at the COW pin, but it must also pull C1 up to the new voltage level. This load is unfairly heavy, too much for the COMP pin's control amp, and the output will be slow to regulate. It's possible to avoid this problem by adding a 470 k-ohm resistor from the top of C, to the input supply. This resistor pulls CI up to a high level, where it cannot interfere with operation after start-up.
Even after you add the resistor, this circuit would be in trouble-specially if the input power shuts off briefly. It takes many seconds to discharge C1, a 5 uF capacitor, and after a brief power outage no soft-start capability would be available. A good fix is to install a diode across the 470 k-ohm resistor to discharge C1 quickly when the input supply voltage drops. This gives you a chance for soft-start on the restart.
I'm not saying that this circuit is a good worst-case design-you'd have to prove that with engineering and tests. But it's not as bad as it was. Meanwhile, we have at least added these components to our LM3524 data sheet.
Toys Illustrate Some Basics
My first encounter with a start-up circuit happened at a pretty young age. I remember the old toy that resided in a box with a big ON-OFF toggle switch on its front. When your curiosity led you to flip the switch ON, a motor would begin to whir, the top of the box would rise up, and a mechanical hand would reach out. The hand would throw the switch to the OFF position and then retreat back into the box. The cover would close, and the whirring would cease. What a charming way to represent a startup (or, in this case, a shut-down) function. When I was a kid, I was really impressed by this toy, but later I realized that this "logical sequencer" was an illusion. The ON-OFF switch did trigger some kind of a latch to turn the power ON, but it did not directly provide the OFF function. If it did, the hand would stall immediately after it turned the switch OFF, and would not continue back inside the box. There WJUS a switch to turn the power OFF, but it was functionally inside the box, triggered by the end of the travel of the hand.
The point of the story is that we must make sure not to fool anybody--especially not ourselves-with the start-up circuits we design. When I design an A/D converter, I include a shift-register sequencer to make sure that every necessary task is completed sequentially at the start of conversion, and then again at the end. I have no idea how software engineers write valid programs to make sure no false start-up sequences occur, but I bet some people do it wrong. Some designs still implement the micro processor RESET function with such crude RC timers that the processor gets confused or faked-out and doesn't start correctly. I've heard about those horror stones.
Sometimes people forget to add the diode to discharge the capacitor, and then the reset fails to work after a brief power outage. (The same diode function as D2 in FIG. 2.) Designers also include start-up circuits in linear circuits. For example, current-bias circuits ensure that a small but stable bias current starts the whole circuit going. Then when the start-up circuit is operating properly, another circuit (a "hand") reaches around and shuts off the start-up circuit. When the start-up circuit works properly, it saves power and doesn't waste much die area. Unfortunately, if the start-up circuit is broken or inoperative, the main circuit may still start if the supply voltage jumps up rapidly but may fail to start if the supply ramps up slowly. Once a customer returned a regulator complaining that the device would start if the supply ramped to 20 V in 30 seconds or less, but would not start if it ramped in 36 seconds or more. We checked it out, and the customer was absolutely right. We had to change one mask and add a start-up test to prevent future troubles.
Over 20 years ago, an IC maker designed a micropower IC that didn't have a true DC start-up circuit. The IC was supposed to be started up by the transient rise-rate of the supply voltage. At room temperature, the circuit would always start, no matter how slowly the supply ramped up. But at cold temperatures, the device wouldn't start even with supplies of +15 V, if the supplies came up slowly. Worse yet, if the device was running and you hit one of the power supply busses with certain positive or negative transients in sequence, you could turn the part OFF, and it might never turn ON again. Needless to say, that part never became popular, nor did anything else from that company.
So, let me caution you: Whether your circuit is a loop of sequential logic or an analog loop with positive and negative feedback, be sure to design the start-up circuit carefully. Add a test to make sure a bad part will be rejected, AND build up a few bad parts--units with the start-up circuit broken or disconnected--AND make sure they fail the test. Then leave the test in the flow. Don't drop the test just because nobody has ever seen a part fail. Dropping that test would be courting disaster. Here at National Semiconductor, we've appointed a Czar of Start-up Circuits. He is the repository of all knowledge about circuits that do (and don't) start properly. Since this shy fellow (I shan't give his name) began to reign, the goof-up rate has been cut by many decibels.
1. Dietz, Erroll H., "Reduce Noise in Voltage Regulators," Electronic Design, Dec. 14, 1989. p. 92.