Ever since pioneering device engineer Bob Widlar in 1965 designed the µA709 IC for Fairchild, operational amplifiers have been the analogue-design engineer's equivalent of glue logic. But in those early years, you had to be a competent analogue designer to apply most op amps with any real success. Early devices required careful frequency compensation and lacked any output short-circuit protection, making them vulnerable during benchtop development. Nonlinear circuit design, such as true rms-to-dc conversion, was especially tricky, requiring complex compensation networks to provide stability.

Ironically, perhaps, today's emphasis on digital systems shifts such computational duties from continuous-time to clocked-circuit operation, but systems engineers require more op amps than ever before to bridge the analogue-to-digital divide. And with fewer and fewer analogue specialists in the industry, many logic designers now face the responsibility for making these real-world connections. Partially in response to this trend, semiconductor-device designers continuously introduce new generations of parts that not only boast specifications that were unimaginable when the µA709 appeared but are also much simpler to apply.

Just one decade ago, glancing through vendors' catalogue listings might have convinced you that op-amp design had reached something of a plateau, with a device to suit every application. But, as with everything else in this industry, nothing stays the same for long. Two main factors now challenge semiconductor-device and end-user equipment designers alike: the trend toward single-supply operation and the explosive growth in mobile devices. Each of these factors adds its own requirements, but both share the ever-lower power-consumption requirements that contemporary designs demand. Although it's obvious that the ever-growing list of battery-operated electronics, such as cell phones, digital cameras, handheld games, MP3 players, notebook PCs, and PDAs, require the lowest possible power consumption, the trend doesn't stop with consumer durables. Systems engineers are now commissioning environmental-monitoring and industrial-process-control systems that consume a quarter of the current that their counterparts required just five years ago.

Single-supply operation now dominates op-amp applications for several reasons. First, it's convenient; you no longer have to design and accommodate multiple power supplies. Just lowering the supply voltage from the traditional ±15V to, say, 5V helps you to conserve energy and minimise power dissipation, which in turn reduces your cooling requirements. You also minimise power losses by eliminating multiple voltage conversions. Second, if you use the same supply level that logic such as ADCs require, you can neatly match your signal-conditioning circuitry to the ADC's input level. But single-supply operation presents a number of challenges, starting with the op amp's natural desire to operate from split supplies.

Because op amps are linear devices, they operate in four quadrants that take the math form of y=Mx+C, where M is the circuit's gain term, and C is a constant that represents an offset value. To provide the equal positive and negative excursions that maximise a circuit's dynamic range, an op amp's input-signal level ideally centres between its supply rails. Also, the amplifier's internal circuitry invariably works best midway between the supply rails, with traditional designs becoming nonlinear as signal voltages approach the supplies. Together, these behavioural features require you to implement a virtual ground that's typically halfway between the supply rails.

This virtual ground often crucially determines overall circuit performance. A crude way to generate a midrail reference uses two resistors as a divider, with a capacitor to ground to help attenuate power-supply noise. And therein lies the first problem: Any remaining power-supply noise feeds directly into the reference input and adds to input-signal noise. In dc terms, the circuit's output impedance equals the resistor values in parallel, and ac response is poor, introducing frequency-dependent distortion.

It's also difficult to strike a balance between output impedance and power consumption. Assuming equal-value 1-kΩ resistors and a 5V supply, the output impedance is 500Ω for 2.5 mA. This value represents 12.5-mW dissipation in the resistors, and more accrues if you're using a power-supply voltage regulator. It's better to select a low-power voltage-reference IC, but for minimal output impedance and best noise rejection, ensure that this device can source and sink current. Like linear voltage regulators, almost all low-power voltage-reference ICs are designed as current sources. One exception is Linear Technology's LT1118-2.5, which achieves an output impedance of around 0.025Ω and can drive ±800 mA for an I_Q (quiescent current) of some 600 µA.

Lowering the supply voltage intrinsically reduces the op amp's output voltage swing, so the circuit's dynamic range immediately suffers. This consideration can become crucial with designs that must work down to the voltage supplied by a pair of NiCd batteries approaching deep discharge at, say, around 1.8V. Typically, all circuits must accommodate maximum input-signal levels that include transient overdriving without performance degradations, such as distortion due to nonlinear amplifier response.

Some sensors, such as piezoresistive-bridge accelerometers, can be microphonic and generate transient overload levels far exceeding their normal output. In extreme cases, input-signal overdrive levels can cause phase reversal in some op-amp designs—the analogue-circuit equivalent of a digital device's latch-up. And, there's the SNR issue to consider: Following amplification, a few tens of microvolts of input-referred noise can become an unhealthy percentage of a low-voltage supply's headroom. For this reason, fully differential signal processing is becoming more popular; because the noise terms between two op-amp drivers are uncorrelated, you immediately improve SNR by 3 dB.

Among other attributes, an ideal op amp has infinitely high gain and bandwidth, infinitely low output impedance, and zero noise contribution. Unless you're constructing a comparator circuit, negative feedback dictates the circuit's response. First described by Bell Labs communications engineer Harold Black in 1934, the negative-feedback principle applies a fraction of a high-gain amplifier's output back to its input to correct a circuit's transfer-function errors. Thus, you can visualise operation within negative-feedback circuits simply by remembering that—provided that the circuit is operating within its linear region—feedback action drives the voltage between the input terminals to zero. By considering the current flow within the feedback network under given input conditions, mathphobes can escape complex-looking formulas to intuitively arrive at the correct result. But for formal analysis and a collection of single-supply circuit examples, download Texas Instruments' free "Op amps for everyone" design reference, which focuses on application-level issues (Reference 1).

Before you consider application issues, it's essential to appreciate how components work. You can see the functional blocks that comprise a typical op amp by examining the ubiquitous 741 IC (Figure 1). Designed in 1968 by Widlar's successor at Fairchild, Dave Fullagar, the µA741 was the first device that was genuinely easy to apply. By today's standards, the circuit is simple, but its same principles apply to all general-purpose op amps. The design also illustrates several factors that impact single-supply and low-voltage operation.

The input terminals are the bases of a pair of npn bipolar-junction transistors, Q₁ and Q₂, in an emitter-follower configuration. To provide isolation between the input signals and the stages that follow, the input pair immediately feeds two pnp transistors, Q₃ and Q₄, which form a common-base amplifier. This strategy decouples the input terminals from the following stages and prevents signal feedback at high frequencies. Transistors Q₅ and Q₆ form a current mirror that supplies constant bias current to the input stage and allows maximum headroom to the positive supply rail in the presence of large common-mode voltages. The Q₇/Q₈ pair load the stage and provide the differential-to-single-ended conversion that's necessary to drive the second gain stage.

Bias for Q₇ and Q₈ comes from Q₉, which maintains the voltage across Q₇'s collector-emitter at close to two diode drops. Analogue-circuit guru John Pickering, principal of instrumentation consultancy Metron Designs, explains: "Imagine that there are no volts across the 50-kΩ resistor, and Q₇/Q₈ are off. Then, the voltage at Q₇'s collector rises, due to current from Q₃. Transistor Q₉ now 'emitter-follows' it to just above 2V_BE, when its emitter supplies enough current to turn Q₇ and Q₈ on—and also supply the 50-kΩ resistor." Pickering notes that Q₇'s collector-base path via Q₉ is a negative-feedback loop that causes the circuit to stabilise with Q₇'s collector at about 1.3V above the negative rail. Also, Q₇'s collector is loaded only by Q₉'s base current. According to Pickering, adding Q₉ is better than diode-connecting Q₇ as in a normal current mirror because Q₉ supplies the base current for both Q₇ and Q_8, rather than from just one side of the mirror. Thus, only Q₉'s base-current requirement compromises the matching between the mirror devices.

One of the key challenges for op-amp designers is determining how to level-shift between successive gain stages and still maintain sufficient headroom between the supply rails. The 741 makes extensive use of Widlar's current-mirror circuits to shift and scale the biasing throughout its three stages. Pickering expands this concept, explaining that the 39-kΩ resistor together with Q₁₀, Q₁₁, and their associated mirrors, set the op amp's internal bias-current levels. This circuit defines a current in Q₁₂ that pulls the bases of Q₃/Q₄ until there's sufficient current in Q₁/Q₂ to drive the mirror Q₅/Q₆ to supply the current to Q₁₂. That is, a dc bias-current-control loop maintains its operation over Q₁/Q₂'s full common-mode-voltage range.

Working as a floating reference, Q₁₃ and its base resistors form a VBE multiplier; neglecting Q₁₃'s base current, the current through the 7.5-kΩ resistor must also flow through the 4.5-kΩ resistor. Approximately 1V appears across Q₁₃'s collector-emitter, and this voltage controls the crossover point between Q₁₄ and Q₁₅ conducting. For output-stage protection, Q₁₆ provides short-circuit current limiting for Q₁₄ by diverting current from Q₁₄'s base when an overload is present. The protection for Q₁₅ is subtler in that Q₁₇ samples the output-stage current and then diverts Q₁₈'s drive current to the negative supply rail. Also, the integral-frequency-compensation capacitor, C₁, guarantees that the device is stable down to unity gain—a novelty at the time and a huge factor in popularising op-amp applications.

You can see that a 741-style design has problems accommodating common-mode input voltages that approach the supply rails. In either the positive or the negative directions, the Q₅ current-source and Q₇/Q₈ current sinks require ≥2V of compliance voltage to work well. The data sheet reflects this fact, stating that the input common-mode voltage range is typically ±13V for ±15V supplies. Also, the output voltage can't reach the supply rails, because each npn or pnp member of the output pair cuts off within a volt or so of the respective rail.

To improve performance in the single-supply circuits that automotive applications demand, National Semiconductor in 1974 introduced the LM324 single-supply rail op amp. With four op amps per package and able to operate from a 5V supply, the LM324 suits common logic families and is still the world's best-selling op amp. Uniquely for its time, the LM324's input common-mode range includes ground, and its output can also reach this level. Plus, the differential input-voltage range equals the supply voltage. These abilities rely on the extensive use of pnp transistors that were previously unpopular with op-amp designers because of their inferior performance and greater die area compared with npn devices (Figure 2). But the output level still limits at around two diode drops below the positive supply rail, intrinsically reducing output dynamic range. And for applications such as driving ADCs, output dynamic range is typically more important than an op amp's input common-mode range.

To get the maximum code-output range from a conventional ADC, you need to drive its input between ground and its supply rail. This dynamic-range consideration has led to today's generation of op amps featuring RRO (rail-to-rail-output)-level capabilities. For bipolar-junction-transistor designs, RRO op-amps reverse the 741's emitter-follower outputs to substitute a pnp device to the supply rail and its npn complement to ground. In this way, the outputs can swing within millivolts of either rail. But the device often pays a price that is not obvious in the data-sheet specs: Because the output terminal now connects to the collectors of the output-driver transistors, the device's output impedance becomes much higher than it is with a traditional emitter-follower amplifier.

Bill Gross, general manager at Linear Technology in Milpitas, CA, observes that customers who have been successfully designing single-supply circuits often encounter problems when they apply RRO devices for the first time: "For a given power consumption, RRO op amps have less drive and less bandwidth than conventional devices." Gross explains that Linear Technology always specifies its RRO devices as working into specific loads. He also notes that if an RRO device has an output impedance of 250Ω, and you connect a 1-kΩ load, the resulting gain of four makes it harder to close the loop and provide unconditional stability.

High-speed RRO devices, such as video amplifiers, can be prone to oscillation—sometimes even when you apply a scope probe's few picofarads of stray capacitance to the op amp's output terminal. For this reason, Linear Technology endeavours to make all its op amps capable of driving at least 50 pF. You might also consider adding a decoupling resistor between the op amp's output and a test point, but this extra load in series with a 10× scope probe introduces a proportional gain error. Interestingly, techniques that Linear Technology uses to minimise output impedance in the LT1800 IC include adding a pair of capacitors (C₁ and C₂) that provide local negative feedback to the output driver's bases (Figure 3). Built using bipolar process technology and specified for 3 and 5V operation, the LT1800 has a typical 80-MHz GBW (gain bandwidth) for 2 mA of I_Q; it is stable at unity gain for maximum loads of 75 pF.

Designers consider this gain-bandwidth-per-milliamp value a measure of op-amp efficiency. And, not surprisingly, the relationship between bandwidth and power consumption is a prime op-amp selection parameter (see sidebar "Simplify op-amp selection"). For lowest power applications, Martin Giles, amplifier-applications manager at National Semiconductor's Santa Clara, CA, facility, recognises a trend for op amps to include shutdown circuitry that minimises I_Q when the device is not in use. But, because the shutdown circuitry works by disabling various parts of the op amp's biasing circuits, a trade-off exists between shutdown current and wake-up speed. Many designers overlook this consideration. "When a vendor quotes you a shutdown current figure," Giles explains, "always ask how long the device takes to restart; the answer may surprise you." For example, National Semiconductor's LMH6647 is a 55-MHz-GBW device that suits 2.7V operation and draws approximately 650 µA; the LMV981 has a 1.4-MHz GBW and consumes about 100 µA using a 1.8V supply. Both devices feature shutdown inputs that typically reduce I_Q to 15 and 0.156 µA, respectively. But the LMV981 takes a leisurely 12.5 µsec to wake up; the LMH6647 is working within 250 nsec.

Like the LT1800, the LMH6647 and LMV981 follow the growing trend for low-voltage op amps to include RRI (rail-to-rail-input) capabilities as well as RRO capabilities; such devices earn the description "RRIO." Giles notes that National Semiconductor prefers BiCMOS technology for its low-voltage op amps because it's a truly complementary process. Bipolar processes are invariably quieter for input stages and also easier to control at low voltages, but CMOS eases driver-stage design. The main problem is in reducing the threshold voltages for the CMOS output-stage elements. Giles eagerly accepts that the ability to accommodate input voltages down to ground and below is often useful but points out that input-referred operation to the positive supply rail can be questionable, stating that in a normal application, RRI is useful only if you're building a unity-gain buffer.

There's a price to pay for the convenience of RRI operation that's a consequence of the dual-input stage that such amplifiers require (Figure 4). This configuration leads to three operational modes. Between ground and about 1.2V below the positive supply rail, the Q₁/Q₂ pnp pair is active. As the input voltage rises above this level, biasing arrangements start to turn off Q₁/Q₂ and turn on the npn pair, Q₃/Q₄. Thus, there's a transitional period. Eventually, the pnp transistors turn off completely, and the complementary pair provides the first stage of gain. Therefore, over the full input range, the polarity of the input bias current changes. And, because pnp and npn transistors have different gain properties, this current's magnitude also sharply changes, with a plateau during the transitional period.

Apart from introducing additional dc errors that are hard to externally or internally compensate, such stages can generate distortion during the transitions between operating modes. The message is that, for best performance, don't use RRI devices simply because you think they're convenient. Applications that can benefit from RRI inputs include high-side current-sensing applications, in which the sense resistor is at or above the op amp's supply rail. But a better option may be to use a current-sense IC, such as Texas Instruments' INA168. This op-amp based circuit accepts common-mode input voltages reaching 60V and consumes just 25 µA.

You may at times consider alternatives to conventional op amps. One such example is when you're interfacing with Wheatstone-bridge circuits, such as accelerometers, thermocouple bridges, and pressure-sensor ICs. The normal connection employs an instrumentation amplifier, which you can construct from three op amps and a handful of precision resistors. For adequate common-mode rejection, you must often match these resistors to much less than 1%. But you can now get low-power, low-voltage instrumentation amps that integrate these elements and avoid component-matching difficulties. Typically, you program such amplifiers to operate at gains of 1 to 1000 with a single resistor. The micropower AD627 from Analog Devices is a classic three-amplifier design that provides RRO operation for a maximum I_Q of 85 µA. The device accommodates supplies of ±18V down to just 2V to suit power-reduction schemes in everything from process-control systems to battery-powered instruments. But with performance specified for 20- and 100-kΩ loads, you'll need a buffer to drive typical ADCs. Also notice that the minimum gain figure is five, but this restriction is unlikely to challenge most applications.

There are alternatives to the three-op-amp configuration that some vendors employ to build instrumentation amplifiers. Optimised for dc accuracy within a 400-Hz bandwidth, Linear Technology's LTC2053 boasts a maximum V_OS (offset voltage) of just 10 µV and an offset-drift figure of 50 nV/°C. The device consumes a maximum I_Q of 1.1 mA from supplies that can span ±5.5 to 2.7V. Features include RRIO operation that's specified for loads down to 2 kΩ. The device achieves this performance by using a switched-capacitor input architecture followed by a zero-drift op amp (Figure 5).

"Zero-drift op amp" is almost always a pseudonym for "chopper-stabilised," and the LTC2053 is no exception. It employs switched-capacitor techniques to synchronously modulate and demodulate the input level and nullify dc errors (Reference 2). The LTC2053's input circuit level-shifts the dc common-mode input voltage to its reference pin's level by sampling the input signal on capacitor C_S and transferring it to C_H. The sampling frequency is 2.5 kHz, and the capacitors are both 1 nF, so you must consider settling time. Provided that the input-source resistance is less than 10 kΩ and gain is less than 100, the switched-capacitor input's characteristics dominate, and the IC settles to n bits of accuracy within about n clock cycles.

At high source-resistance values, a step change in input voltage allows too little time for the capacitors to charge fully within one cycle, so settling time suffers. The capacitors also require an input-charging current that ideally reaches zero for dc inputs. But parasitic capacitors disturb this charge, so the input-charging current decays exponentially during each sampling period, with a time constant that's the product of source resistance and C_S. Notice that if the input voltage across C_S settles within one clock period (400 µsec), the input architecture eliminates errors due to source resistance or source-resistance mismatches.

The INA326 from Texas Instruments demonstrates another innovative approach to achieving the high common-mode rejection that characterises instrumentation amplifiers. To avoid the resistor-matching issues that plague traditional topologies, the INA326 converts input voltages to currents (Figure 6). The outputs from op amps A₁ and A₂ impress the differential input voltage across R₁, which in combination with R₂ sets the amp's gain at 0.1 to 10,000. Both op amps include current mirrors in series with their supply lines, and these supply lines operate from charge pumps that lift them above the external power-supply rails. A₂'s current-mirror adds the current in R₁ to its duplicate from A₁ to double the output value that flows through R₂, providing a gain equal to (2×R₂/R₁). This approach achieves 110-dB common-mode rejection, a maximum V_OS of 125 µV, and a V_OS drift of 1 µV/°C. The charge-pump arrangement permits RRI operation that extends 20 mV beyond the supplies. Other statistics include 1-kHz GBW, single-rail 2.7 to 5V operation, and drive to within 10 mV of either rail into a 10-kΩ load. The chip's I_Q is typically 2.4 mA; the similar INA327 adds a shutdown pin that reduces off-state current drain to a maximum of 5 µA.