In the beginning, there was TTL. And TTL ran from 5V power supplies. Design engineers recognized the advantages of operating most of their circuits from a single standard supply voltage. And the industry adopted 5V. And it was good. Then it came to pass that the new semiconductor processes that produced parts such as the latest high-performance µPs yielded devices that wouldn’t run at 5V. The new parts required lower voltages. The first of these was 3.3V...

If you’re designing portable or battery-powered products, this situation confronts you with a Hobson’s choice: Omit functions that can differentiate your new products from competitors’ or add power-conversion devices in addition to the logic ICs that implement the new features. Certain functions just aren’t available in parts that run at 5V. TTL-compatible parts that use 3.3V supplies can drive standard 5V logic, but can’t run from 5V. Many 5V parts can’t run (or don’t run properly) from lower voltages.

Just as engineers began to think they might devise ways to cope with these incompatibilities, rumors of parts with still lower supply requirements—3, 2.9, 2.7, 2.5, and even 2.2V—started to filter out of Silicon Valley. If designing with two logic-supply voltages is inconvenient, designing with three or more (or picking and choosing devices to keep the number of logic supplies down to two) can be really painful. Reliably and economically generating multiple supply voltages in a small space and with a minimum of wasted power is only part of designing for a mixed-voltage world. It is, however, the part we’re addressing here. We’ll leave for another day the subject of interfacing between logic ICs that operate from different voltages.

Problems become opportunities

When one group of IC companies creates a problem, other IC vendors are quick to see an opportunity. Companies such as Linear Technology, Maxim, and Micro Linear already offer ICs that generate both 5 and 3.3V (see Table 1). These companies are likely eventually to produce products that address the proliferation of logic-supply voltages below 3.3V. Right now, though, only a few ICs create supplies below 3.3V. Some resemble the parts that produce 3.3V. Many of those are in surface-mount packages—some thin enough to mount in 5-mm-thick Type I PCMCIA (Personal Computer Memory Card International Association) cards. A few of the 3.3V parts even provide outputs programmable to 12, 5, 3.3, or 0V at 60 mA, as the PCMCIA standard envisions.

System engineers can readily apply these parts; the data sheets—unlike those of many power ICs—don’t need experienced power-supply engineers to interpret them. The ICs do, however, require external components. So turning them into working supplies requires some effort. Even though combining the ICs with capacitors, inductors, resistors, and FET switches is the most cost-effective approach for such high-volume products as cellular phones, notebook PCs, and personal digital assistants, it sometimes isn’t the way to go. For products built in smaller quantities, in which holding down the development cost is more important than limiting product cost, modular or hybrid dc/dc converters can make sense. Table 1 lists several such converters.

When you first think about generating multiple logic-supply voltages, several alternatives quickly come to mind (see box, "Some power architectures for mixed-logic-voltage systems"). In portable equipment, especially in battery-powered products, the use of a central multiple-output supply, such as an open-frame switcher, can be inappropriate. This scheme finds wide use in medium-sized products that aren’t intended to be portable.

In products that use batteries, a common approach is to locate the batteries in the main product and charge them from an external ac-to-dc supply. Some products require you to remove the batteries for charging. In others, you can charge the batteries in place. In either case, a single-output ac/dc supply charges the batteries, and the unregulated battery output drives the power-conversion devices that produce regulated voltages to drive the logic ICs. If the product uses multiple supply voltages, this scheme is quite similar to the distributed-power architectures used in some larger products (Ref 1).

Portable electronic equipment most often uses NiCd, NiMH, alkaline, or sealed lead-acid batteries, all of which exhibit a large ratio of voltage at full charge to voltage just before they need recharging (or for nonrechargeable batteries, the voltage at the end of life). For most types, this ratio is at least 1.5-to-1. Therefore, a power-conversion device that produces a logic-supply voltage from a battery must operate from a wide range of input voltages. As Ref 2 explains, the design of the conversion circuits is much more straightforward if the battery voltage is always higher (or always lower) than the regulated output voltage.

Although Table 1 lists some devices that can manufacture 3.3 or 5V from voltages as low as 1.2V and so are suitable for use with a single cell (at least a single lead-acid cell; such cells produce approximately 1.65 to 2.35V), you probably want your battery to comprise several cells in series. Just before they need recharging, NiCd and NiMH cells each produce about 0.95V. Hence, six cells in series produce about 5.7V. Several of the ICs in Table 1 operate with inputs as low as 5.5 or 5.4V.

At the other end of the scale, a fresh alkaline cell produces somewhat more than 1.5V. Six alkaline cells in series produce 10V or even a bit more. The ICs that produce 5 and 3.3V with inputs of 5.5 or 5.4V are quite comfortable with 10V inputs. In fact, some of them accept 30V. That’s more than enough to work with the 15V you can get from a fully charged battery consisting of six lead-acid cells.

Several low-dropout linear regulators can produce 3.3 or 5V from an input that ranges from 5.4 to 10V. (V_DROPOUT is the minimum input-to-output voltage for which a regulator produces a regulated output. If the voltage across the IC is less than V_DROPOUT, the IC does not regulate; changes in the input voltage appear substantially unattenuated at the output.) Nearly any linear regulator—low-dropout or not—can produce 5V from an input that varies between 10 and 15V, the range of voltages supplied by six lead-acid cells (automotive batteries contain six lead-acid cells).

Nevertheless, linear regulators are in most cases poorly suited to producing logic-supply voltages directly from battery voltages. The reason is low efficiency—at least when a fresh battery produces a voltage at the higher end of the regulator’s input range. A linear regulator that runs from 15V and delivers 3.3V is 22% efficient. It turns 78% of the power delivered by the battery into heat. Although the regulator’s efficiency increases to as much as 33% as the battery voltage declines, the dissipation may even necessitate expending additional energy on cooling.

For battery-powered applications, switch-mode regulators are the rule. For one thing, their efficiency is nearly always higher than that of linear regulators (full-load efficiencies usually top 70% and often go above 95%). Switch-mode regulators’ efficiency is also, to a first order, independent of input voltage. However, switch-mode regulators are generally more expensive than linear regulators, and they produce EMI, which can be troublesome, especially in applications that involve low-level analog signals.

There are several circumstances in which linear regulators (or switching regulators with a narrow input range) are good choices in battery-powered products. Suppose you are redesigning a product, all of whose logic operates from 5V. In the successor product, most logic will still run from 5V; only one device (or a few devices) will run from a lower voltage. In such cases, you may well want to make the lower voltage from the 5V supply. A low-dropout linear regulator, such as the TC55RP from TelCom—which has a typical V_DROPOUT of 380 mV at 200 mA—or the LT1121 from Linear Technology, easily deliver regulated 3.3V, even if the 5V supply is at 4.75V, the low end of its tolerance.

In this example, the linear regulator’s efficiency is almost 70%. If the 5V supply actually produces 5.25V, the efficiency is still almost 63%.

So here, you must ask whether a switch-mode regulator’s higher efficiency justifies its added cost and complexity.

Another device worth considering in this example is Semtech’s EZ Dropper. EZ Droppers aren’t regulators; they’re 2-terminal linear devices that simulate low-voltage zener diodes. You put them in series with the output of supplies whose output you want to drop, for example, a 5V supply that you want to reduce to 3.3V. (There is no such thing as a true 1.7V zener.) You might ask why you shouldn’t do this job with a pair of forward-biased silicon diodes. The answer is that EZ Droppers offer a more closely controlled forward voltage and a lower series impedance.

As supply voltages decrease, a bit of extra design effort, cost, and complexity will be involved in building switch-mode supplies whose efficiencies match those of today’s 5V-output units. Fortunately, the technology for producing even more efficient switch-mode regulators is in hand: Instead of using silicon-junction or Schottky diodes as rectifiers, power-supply designers will use synchronous rectifiers. The technology has existed for decades but has only recently been applied to power conversion. A synchronous rectifier (often, a bipolar transistor) has a forward voltage that is a small fraction of the 700 mV to 1.2V of a typical silicon-junction diode or the 400 to 700 mV of a typical Schottky rectifier.

Decreasing supply voltages increase the significance of supply output tolerances, which so far haven’t been very important in low-power applications. Devices that operate from 5V are usually unfazed by supply voltages that differ from 5V by 0.25V. If the supply voltage is 2.5V, however, 0.25V represents a 10% tolerance. More than likely, the limits on 2.5V parts’ supply voltage will be ±0.1V. The supply’s open-circuit voltage tolerance is only a part of 5V ICs’ allowable ±0.25V supply-voltage variation. Part of the allowable variation represents the IR drop produced by the flow of load current through the supply’s output impedance; IR drops across pc traces also use up part of the 0.25V.

Although low-voltage versions of 5V parts consume less power than the original parts did, the low-voltage parts don’t draw much less current. And there is a real possibility that future high-performance low-voltage parts will draw more current (though less power) than their 5V counterparts did. Given this scenario, the idea of tighter tolerances on the voltage applied to the ICs’ supply terminals is a cause for concern. Having to hold the power-supply-output and pc-trace impedances to less than half of currently permissible values is not a pleasant prospect.

One vendor that boasts of its ability to produce parts with a tight output-voltage tolerance is Power Trends. The packaging technology of the company’s original dc/dc converters won Power Trends EDN’s Innovation Award for power products a few years ago. Some of the parts are hardly bigger than a TO-220 transistor. Yet because of their high efficiency, they can deliver substantially more power than a linear regulator in a similar package. The automated manufacturing process that produces the parts permits laser trimming of output voltages to a 1% tolerance.

Isolated dc/dc converters can help to mitigate the effects of IR drops in ground lines and ground planes. If you dedicate an isolated converter to powering one IC or a small cluster of ICs, you can limit the current that must flow long distances in ground lines. IR drops in ground lines then use up less of the ICs’ supply-voltage tolerance. Nonisolated power-conversion devices can’t accomplish the same thing. Table 1 lists several small isolated converters.

Another potential problem is dynamic power-management schemes, which cause CPU clock rates to change rapidly. As the clock rates vary, so does the supply current. Without effective decoupling close to the devices whose supply current varies, rapidly changing supply currents cause nasty voltage spikes on supply lines and ground. These spikes may even be large enough to corrupt data.

Still another problem that may arise from using multiple logic-supply voltages is the need for sequencing of power-supply outputs. It is conceivable that ICs can suffer damage if a product’s several power supplies power up in the wrong order. Some of the power-conversion products that Table 1 lists have terminals that must be grounded (or ungrounded) before an output can appear. You can only hope that further investigation shows that if all of the logic voltages in a system come up at their own rates, no damage will result. Few users or designers of small systems would welcome the added complexity and unreliability that would result if supply sequencing were proven necessary.

1. Small, Charles H, "Distributed power takes center stage," EDN, April 28, 1994, pg 54.

2. Moore, Bruce D, "Step-up/step-down converters power small portable systems," EDN, February 3, 1994, pg 79.