Click-Picks

Few designers spend time pondering the traits of resistors, capacitors, or other simple passive components. Determine the necessary nominal value, tolerance, and temperature coefficient for each instance in your circuit, and you're pretty much done. Consider the nonelectrical variables-package and pricing-and you're a certifiable good corporate citizen. Yet the lowly switch, the most conceptually simple device of all, requires more careful consideration.

Why a device that models ideally as either a piece of wire or an open circuit should require design attention at all might seem mildly paradoxical. One explanation hinges on the either/or nature of the model and harks back to the first week of your freshman circuits course when the professor introduced the fundamental assumption of linear, time-invariant circuits-an assumption that may have stood with little comment for the remainder of your collegiate experience. Switches, unlike other components, form time-variable topologies in an otherwise-linear domain, expanding the circuit analysis to a series of state-specific case studies connected by often-ungraceful transitions. The nature of these transitions complicates practical models in ways that depend largely on the implementation. For example, mechanical and electromechanical switches are subject to contact bounce, but their control signals tend not to couple into the switched lines, as is the case with solid-state analog switches.

Although realizable devices are not ideal, or even as close as mechanical or electromechanical samples, the advantages of solid-state analog switches have made them increasingly compelling over the many years they've been available. Greatest among the solid-state switches' benefits are size and cost. In extremis, where integrated in sufficiently complex mixed-signal circuits, the size and cost of the switches vanish within the larger structures. By comparison, mechanical switches and electromechanical relays are large and expensive, and they suffer from wear and failure modes that render them less reliable than their solid-state counterparts in many large-volume applications. In many examples, semiconductor switches enable functions that are impractical by other means. For microcontroller-based user interfaces-long the dominant approach for electronic products in nearly every sector-solid-state analog switches provide a logic-level-compatible control interface and require little activation energy. The effect on user-interface designs has been notable in applications as disparate as dashboard-automotive electronics, portable test-and-measurement equipment, process-control instru-ments, and wireless consumer devices.

Model behavior
Solid-state switches are available for controlling power feeds, digital buses, or analog signals. Those for analog-signal switching are most often formed of a complementary MOSFET pair (Figure 1). The control path typically includes an inverter to provide the complementary drive that the NMOS- and the PMOS-switch devices require.
The two switch devices have parasitic capacitances that couple small samples of the control inputs to the signal line (figures 2a and b). Note that the diagrams stress the device symmetry near zero signal and may not correspond to Spice models or other representations that are optimized for other purposes. Because the complementary control signals give rise to opposite impulses coupling to the channels, one device largely cancels the charge that the other device injects into the signal line. The two devices, however, require different dimensions to exhibit like on-state behaviors, which upsets the balance and leaves the net coupling term, CD (Figure 2c).

Charge-injection tolerance differs by application. For example, synchronized video switches change state during a retrace interval, coordinating the signal disturbance so as to leave no visible artifacts. Other switching environments in sectors as varied as medical analysis and imaging, test and measurement, and professional audio, have little tolerance for transient switching effects. A sample-and-hold amplifier, as a subcircuit example, captures a fraction-typically half-of the hold switch's injected charge on the hold capacitor where it appears as a pedestal-offset error. In audio applications, charge injection causes "clicks" and "pops" as the charge causes a rapid shift in the switched node's potential. For such sensitive uses, switch designers add suitably scaled dummy-FETs or compen-sating capacitors, CC, and connect them to the appropriate drive phase to "buck out" the charge injection modeled by CD.




Fig 1: A complementary MOSFET pair (MN1 and MP1) and a control-signal inverter (MN2 and MP2) (a) implement an ideal analog semiconductor switch (b).





Fig 2: Charge injection derives from a portion of the control signal that couples capacitively to the signal path (a). The parasitics for the N- and P-channel FETS only partially cancel (b), but manufacturers can add compensating devices, CC, to cancel the residual, CD (c).


An analog switch also exhibits several parasitic terms associated with its static states. Far from constant, the channel resistance, RON, is a nonmonotonic function of the channel and gate-drive voltages (Figure 3, Reference 1). The contradictorily named RON flatness measures the difference between the maximum and the minimum resistance over the stated range of operating conditions. Typical flatness values range from one-tenth to one-third of the maximum channel resistance, though in low-resistance switches, the ratio may be poorer than one-half. FET-channel resistances exhibit a strongly positive temperature coefficient. Keep these three traits in mind when evaluating how your circuit responds to the absolute value and changes in the on-resistance.




Fig 3: The channel resistance, RON, in MOSFET analog switches (a) is a nonmonotonic function of the signal voltage and supply bias (b).


Applications that operate on video or RF signals have sensitivities to two other parasitic terms (Figure 4). At high frequencies, the on-resistance alone cannot model the switch's conduction path loss. CS models a distributed shunt stray extending from the conduction channel and the source and drain tubs to the substrate-an ac ground. The shunt term loads the combination of the channel resistance and the signal's source impedance. The resultant insertion loss depends on the signal frequency and impedance.




Fig 4; Substrate and source-drain parasitics affect a switch's ac performance.


The source-drain capacitance, CSD, and the switch's load impedance determine the high frequency off-state isolation. At low frequencies, leakage currents dominate the off-state behavior.

Solid switches
It's tempting to think about individual applications as if they made homogenous demands on switch technologies. With such a view, your design focus falls to the switch form-poles and throws-and the available amplitudes for drive signals. But, as the models suggest, a product can make a variety of demands on its switching devices, as it does on a more familiarly parametric analog component, such as an op amp.

Some portable entertainment, communication, and medical devices use switches in the input and signal-processing paths for high-impedance microphones and sensors. Small switches with moderate channel resistances suit these circuits. Normally open or closed SPST switches, such as Intersil's ISL43110 and ISL43111 offer a maximum 40½ RON over temperature with a 3V supply. The maximum RON flatness under those conditions is 7½. The switch holds its worst-case charge injection to 10 pC at zero signal amplitude and source impedance.
For higher speed signals on 50½ lines, the 43110 and 11 offer a fairly flat response with a 200MHz, Ð3dB point. Off isolation in such applications is typically better than 90dB at 100kHz. The worst-case leakage currents are 10nA per end in the off-state and 20nA total in the on-state. Intersil specifies these switches at 3.3, 5, and 12V. As you should expect of MOSFET switches, the switching times and channel resistance drop and the flatness flattens with increasing supply voltage. The unipolar ISL43110 and ISL43111 operate from Ð40 to +85¡c and cost 51 cents (100) in SOT23-5 packages and 55 cents in SOIC-8 packages.

For applications with bipolar supplies, Intersil offers ISL43112 and ISL43113 with similar specs. The bipolar parts enjoy slightly lower off-state leakage currents and slightly worse flatness but are otherwise similar to the other family members. They cost 57 and 61 cents (100) for the SOT23-5 and SOIC-8, respectively.

Intersil is far from alone in this market. On Semiconductor specifies its NLAST4599 SPDT switch at 2.5, 3.0, 4.5, and 5.5V and over a Ð55 to +25, +85, and +125¡c temperature range. At 3V supplies and 85¡c operating-temperature limit, the NLAST4599 offers a maximum 50½ on-resistance with 4½ flatness.

Double-throw switches must specify their break-and-make characteristics. Most uses, such as source and load selectors, require break-before-make throws. Switchable gain or filter stages, though, often benefit from a make-before-break characteristic to prevent amplifiers from operating open-loop even for brief intervals and to minimize glitching.

Circuits that allow comparisons between signals, common in medical, industrial, and test-and-measurement equipment, may provide the means for compensating for a switch's channel resistance but may exhibit a sensitivity to the difference between the channel resistances, DRON, associated with the two throws of the SPDT device.
On's 15-cent (10,000) SPDT is a break-before-make type and guarantees a maximum 2½ DRON. The manufacturer offers only a typical charge injection, 1.5 pC at 3V supply, but in many quasistatic applications that do not require the full subsystem performance during a switching interval, you may need not design to a maximum charge-injection spec. The switch, available in either SOT23-6 or SC70-6 packages, breaks in 15nsec or less; make time is no more than 25nsec, and the manufacturer guarantees a minimum 1nsec underlap time.

Fairchild offers an unusual configuration in its FSA3357 low-voltage SP3T switch that can save you significant board space when you need a signal-input selector that handles more than two inputs, as is common in many portable devices. Consider a cell phone, which has an earpiece speaker, a ring enunciator, and a headset jack. You can route signals to the three loads with a pair of DPDT switches, but that approach has double the channel resistance of and requires greater board space than a single triple-throw device.

The switch offers 7½ channels with 4.5V rails and operates to 1.65V supplies by which point the on-resistance rises to 20½ at zero signal. The switch's channel resistance typically exhibits a subohm matching error. The manufacturer reports the switch's on-resistance flatness as a typical over the full common-mode range of the part, but provides no characterization curves for that interval. Though this approach may give you a conservative expectation of the part's capability with the apparent convenience of a single value representation, it makes comparisons with other manufacturer's products somewhat more challenging.

The 3357 rolls off to Ð3dB at 250MHz. Typical off-isolation and crosstalk are Ð58 and Ð60dB, respectively, both at 10 MHz with 50½ terminations. The data sheet omits characteristic curves for ac parameters, but generic switch models suggest that you can use 20dB/decade as a rule of thumb for off-isolation and crosstalk at frequencies other than the test point with both parameters decaying with increasing frequency. You can also calculate the coupling capacitances those figures imply and recast them for impedances other than 50½. You can also use a single-pole roll-off model (in the poles-and-zeros as opposed to the poles-and-throws sense), which gives an amplitude error of less than 1% at a frequency one decade below the Ð3dB point.

The 51-cent (1000) switch provides two logic inputs to select the throw. The 0,0 state opens all throws-a feature you can use as a mute position when employing the switch as an input or an output selector or as a fourth state if you use the triple-throw device to select gains or filter characteristics (Figure 5). Fairchild packages the FSA3357 in its US8 package, which uses roughly three-quarters of the board area that an SOT23-8 requires.




Fig 5: An SP3T switch, such as Fairchild's FSA3357, makes space-efficient signal selectors (a) or gain switches (b).


Pericom packages a single SPDT in either a TDFN-6 or an SC70-6 and a dual SPDT in a 133mm TDFN-12, both with dc characteristics similar to the Fairchild SP3T. The PI5A3157 single and PI5A3158 dual switches develop 6½ channels with 4.5V supplies. The typical Ð3dB bandwidth is 250MHz. Unfort-unately, the manufacturer neither reports the test frequency at which it specifies the typical Ð57dB off-isolation or Ð54dB crosstalk, nor provides any characterization curves that might shed light on the product's behavior. On the other hand, if yours is a low- or moderate-bandwidth application, you might be attracted to the PI5A3157's and 3158's price tags-22 cents (1000) for the single and 40 cents (1000) for the dual-and their smaller-than-most packages.

Vishay Siliconix has also developed packaging technology sufficient to fit an SPDT into a footprint little more than 1.531.0mm. Unlike Pericom's TDFN-6 package, Siliconix packages its DG3000 in a BGA-6 that the vendor refers to as a micro foot package, which it also uses for an associated line of low-on-resistance power MOSFETs. You can operate the DG3000 on 1.8 to 5.5V supplies. The manufacturer characterizes the switch at 2, 3, and 5V, a range over which the device develops channel resistances of 22.5, 4.2, and 2.8½. The 47-cent (100,000) switch's typical ac characteristics, Ð61dB off-isolation and Ð67dB crosstalk, apply at 1MHz with 50½ terminations.
Not every product handles little signals that fit neatly between 3, 5, or even 7V rails. If your signal traces come with shock-hazard warning labels, the HV20220 from Supertex may be the switch for you. The 20220 is an eight-channel analog switch that can operate on signals as large as 200V p-p. Supertex gives the maximum on-resistances under a variety of signal currents, supply voltages, and temperatures -most falling into the upper 20s and lower 30s of ohms. Worst-case switch-to-switch channel-resistance matching is 20%, measured with 5mA signal current and ±100V supplies.

You control the 20220's eight switches through a shift register that also provides for readback. The logic subcircuits operate from a 4.5 to 13.2V supply. The switches typically inject several hundred picocoulombs, depending on operating conditions. The manufacturer provides no charge-injection maximums. Supertex provides the $17.14 (1000) octal high-voltage switch in a number of packages, including DIP-28, PLCC-28, TQFT-48, µBGA, and bare die.

Low-Z
Channel resistances as low as 1½ are still too high for high-current applications with low-impedance loads. MOSFET power switches are readily available with on-resistances less than 100m½, and some recent examples have less than 10m½ on-resistances, but manufacturers neither design nor characterize those devices as-analog switches. Personal-entertainment, communication, and information devices, which often use 8 or 4½ speakers, make familiar examples. Switching the speaker amplifiers between internal and external speakers can result in excess power dissipation if the switch's channel resistance isn't a small fraction of the load impedance.

Analog Devices offers the ADG819 for such purposes. The device is available in an SOT23-6; µOIC-8; or six-bump, 1.132.2mm µSP. Its maximum 0.8½ channel-resistance spec applies to 5V supplies across the automotive-temperature range of Ð40 to +125¡C. The 93-cent (1000) switch's on-resistance flatness holds to 0.25½.

The FET dimensions necessary to limit the channel resistance to less than 1½ tend to result in greater charge injection than smaller parts with larger on-resistances. But, all other things being equal, the lower impedance loads better tolerate charge injection than their higher impedance counterparts because a given charge impulse develops a smaller voltage transient. Consequently, you need to adjust your assessment of charge-injection specs to the switched impedance and your circuit's sensitivity to switching disturbances. The ADG819 typically injects 20 pC- number few of its competing subohm analog switches match.

The internode parasitic capacitances are also proportional to size of the FETs. The off-isolation and crosstalk specs, Ð71 and Ð72dB, respectively, at 100kHz, reflect this relationship. The 819 implements break-before-make switching with a 5nsec make delay time. A similar switch, the ADG820, provides make-before-break switching with 15nsec overlap.

Maxim's MAX4684 integrates a pair of SPDT switches in a 1.532mm µSP and operates on 1.8 to 5.5V supplies. It provides a maximum 0.5½ on-resistance over temperature with a 3V supply bias. The $1.15 (1000) switch typically injects 200 pC. Maxim specifies the Ð64dB isolation and Ð68dB crosstalk at 100kHz.

Better blockers
Off-isolation and insertion-loss performance challenge switch designers as the market for solid-state RF switches grows both in size and extent. To keep a switch's low-frequency insertion loss down to, say, 0.5dB on a line with 50½ matched impedances at the source and load, the channel resistance is limited to less than 6½. As you increase the signal frequency, parasitic capacitive terms start to take over the off-state behavior. To make a switch with a meaningful off-state at frequencies greater than several hundred megahertz, switch designers leave the familiar switch topology and instead employ one that provides a means of shunting the off throw and its associated stray signal to ground (Figure 6).





Fig 6: Semiconductor RF switches ground their open throws to shunt crosstalk and enhance off-isolation. You can buy switches of this type with or without integrated terminating resistors.


The ADG919 SPDT from Analog Devices provides an insertion loss of 0.5dB at dc, rising to only 0.8dB at 1GHz. In the off-state, the $1.07 (1000) switch provides 43dB isolation. The CSP (chip-scale package) allows you to fit this device in designs for space-constrained wireless applications, such as an antenna switch, or a band selector in dual-band applications. A similar part, the ADG918, dispenses with the on-chip termination resistors for applications that provide termination elsewhere.

Peregrine Semiconductor specifies its PE4246 SPST from dc to 5GHz. At the upper limit of that range, the switch's maximum insertion loss is 1.8dB, and the minimum isolation is 40dB. Like the ADG919, the $1.98 (10,000) PE4246 is available as the PE4245 without termination resistors. Peregrine also makes a 75½ switch for CATV, operating on signals from dc to 1.3GHz. Its maximum insertion loss is 0.65dB to 50MHz, rising to 1dB at 1GHz. The minimum isolation figures are 85 and 60dB, respectively, for the same frequencies. Both the PE4270 and the unterminated PE4271 cost $1.32 (10,000) and fit into a 333mm MLPM-6.

Manufacturers of solid-state RF switches also often characterize their products for return loss, 1dB compression point, and second- and third-order intercepts.

Author Information
You can reach Technical Editor Joshua Israelsohn at:
1-617-558-4427,
fax 1-617-558-4470,
e-mail jisraelsohn@edn.com

Reference
1. Wynne, John, "RON modulation in CMOS switches and multiplexers," Application note AN-251, Analog

At a glance

The best choice of switch and switch-circuit topology depends on your circuit's impedance, bandwidth, and signal levels and your application's performance demands.

Scale the switch you choose to its environment: Choosing a switch with a lower channel resistance than you need can result in larger switching disturbances than a more appropriately scaled device.

Power and analog switches are not interchangeable even though they may share a few common attributes and specs.

Sidebar: Specs and specsmanship

Every manufacturer wants to paint its product in the most flattering light. Product announcements and the front pages of spec sheets often refer to typical behaviors rather than minima and maxima. From the spec tables, however, you expect to gain a clear and fair picture of a part's capabilities and limitations with few shenanigans.
A couple of legitimate reasons exist, however, for a spec table not to reveal all at first glance. Test time can account for a significant fraction of the final factory cost of simple integrated functions. Testing over a temperature range is particularly costly. Manufacturers may choose to characterize parameters that, in their estimation, offer a poor exchange of test cost and value. An initial product characterization can later form the basis for either correlated measurements or periodic sampling to ensure ongoing performance. Most manufacturers publish their policies on testing and you can check to see if they back all minimum/maximum numbers with 100% testing, correlation to another tested parameter, sample testing, or "guarantee by design." They may also provide useful information about the methods of assurance on individual specs in the oft-unread notes you find at the bottom of the spec table.

A second reason spec tables may mislead-unintentionally or otherwise-is that some specified parameters are strong functions of their operating conditions. If only single-point conditions appear on the spec line, then you need to check the characterization curves that follow to see how the parameter varies over the range of conditions your circuit and its environment will impose. The more forthcoming IC vendors provide ample information in their characterization curves, though you should keep in mind that these representations are usually limited to statistical averages.

A case in point is the charge-injection spec for solid-state analog switches. Many manufacturers publish the maximum charge injection with specific signal-source and -load conditions-usually VT=0V and RT=0½ (Figure A). It is doubtful that any manufacturer intends this stipulated source to mislead. Indeed if they chose a non-zero source, you would likely find understanding the part's behavior more challenging and comparing parts from different manufacturers nearly impossible. Still, this spec comes with two potential "gotchas."

Analog switches are largely sym-metrical devices. You can expect that, for every picocoulomb a switch issues from one I/O pin, it delivers a like charge from the other. The test circuit most manufacturers use necessarily measures one end and, thus, only one-half the actual charge injected into the channel. In many applications, you are interested in transient charges appearing at only one or the other end of the switch. In the earlier example of the sample-and-hold amplifier, the critical node is the hold-capacitor end of the switch. The charge injected back into a low-impedance source has essentially no effect. When assessing your circuit, consider carefully how the charge divides. Depending on source and load impedances, circuit topology, and operating conditions, you may not be assured of an even distribution.

The second caveat concerns the fact that charge injection is a strong function of channel voltage. If your circuit operates its switches off ground, the charge injection can increase substantially over the 0V condition that the spec table lists. Compare the characterization curves with the operating-voltage range your circuit will impose on the switch, and then determine how your circuit will dispose of the resulting charge impulse.

Be aware that some manufacturers choose to report only typical charge-injection values. If your circuit exhibits a sensitivity to charge impulses, you may use this aspect of specsmanship to narrow the field of candidate parts to those that state maxima.