Analog Dialogue 44-0 4, April (201 0) 1 The 20-Bit DAC Is the Easiest Part of a 1-ppm-Accurate Precision Voltage Source By Maurice Egan Introduction A common use of high-resolution digital-to-analog converters (DACs ) is providing controllable precision voltages. Applications for DACs with resolution up to 20 bits, precision up to 1 ppm, and reasonable speed includ e gradient coil control in medical MR I systems; precision dc sources in test and measurement; precision set-point and position control in mass spectrometry and gas chromatography; and beamprobing in scientific applications. Over time, the definit ion of what constitutes a precision integr ated- circuit DAC has changed rapidly as semiconductor processing and on-chip calibration technologies have advanced. Once, high- precision 12-bit DACs were considered to be hard to achieve; in recent years, 16-bit accuracy has become widely available for use in precision medical, instrumentation, and test and measurement applications; and over the horizon, even greater resolutions and accuracies are called for i n control and instrumentation systems. High precision applications for integrated circuits now require 18-bit and 20-bit, 1-ppm-accurate digital-to-analog converters, a performance level previously achieved only with cumbersome, expensive, and slow Kelvin-V arley dividers —the preserve of standards labs and hardly suitable for instr umentation systems in the field. More convenient semiconducto r-based, 1-ppm-accur ate solutions to such requirements using assemblies of IC DACs have been around for some years; but these complex systems use many devices, require frequent calibrations and great care to achieve accuracy, and are both bulky and costly (seeAppendix ). The precision instrumentation market has long needed a simpler, cost-effective DAC that doesn’t require calibration or constant monitoring, is easy to use, and offers guaranteed specifications. A natural evolution from 16-bit and 18-bit monolithic converters—such a DAC is now a reality. The A D5791 1-ppm DA C Advances in semiconductor processing, DAC architecture design, and fast on-chip calibration techniques make possible highly linear, stable, fast-settling digital-to-analog converters that deliver better than 1-ppm relative accuracy, 0.05-ppm/ °C temperature drift, 0.1-ppm p-p noise, better than 1-ppm long- term stability, and 1-MHz throughput. These small, single chip devices have guaranteed speci fications, do not require calibration, and are easy to use. A typical functional diagram for the AD5791 and its companion reference- and output buffers is shown in Figure 1. SPI AD5791 1ppm DAC V REFP V OUT V REFN Figure 1. AD5791 typical operating block diagram. The AD5791single-chip, 20-bit, voltage-output digital-to-analog converter speci fies 1-LSB (least-significant bit) integral nonlinearity(INL) and differential nonlinearity(DNL), m aking it th e world’s first monol ithi c www. analog.com/analogdialogue 1-ppm-accurate digital-to-analog converter (1 LSB at 20 bits is one part in 2 20 = one part in 1,048,576 = 1 ppm). Designed for use in high-precision instrumentation and test and measurement systems, it offers a significant leap in all-around performance compared to other solutions, providing greater levels of accuracy and repeatability in less space and at lower cost, permitting instrumentation applications that previ ously would not have been economically feasible. Its design, shown in Figure 2, feature s precision voltage-mode R-2R architecture, exploits state-of-the-art t hin-film resisto r-matching techniques, and employs on-chip calibration routines to achieve 1-ppm accuracy levels. Because the device is factory cal ibrated and, therefore, requires no run-time calibration routines, its latency is no greater tha n 100 ns, so the AD5791 can be used in waveform- generation applications and fast control loops. V REFPF V OUT V REFPS V REFNF V REFNS 2R S0 R 2R 2R S1 R 2R S13 R 2R E62 2R E61 2R E0 14-BIT R-2R LADDER 6 MSBs DECODED INTO 63 EQUAL SEGMENTS Figure 2. DAC ladder structure. Besides its impressive linearity, the AD5791 combines 9 nV/√Hz noise density, 0.6- μV peak-to-peak noise in the 0.1-Hz to 10-Hz frequency band, 0.05-ppm/ °C temperature drift, and better than 0.1-ppm long-term stability over 1000 hours. A high-voltage device, it operates from dual supplies of up to ±16.5 V . The output voltage span is set by the applied positive and negative reference voltages, VREFPand VREFN, offering a flexible choice of output range. The precision architecture of the AD5791 requires high- performance external amplifiers to buffer the reference source from the 3.4 k ΩDAC resistance and faci litate force-sensing at the reference input pins to ensu re the A D5791’ s 1-ppm linear ity. An output buffer is require d for load driving to unburden t he 3.4 k Ωoutput impedance of the AD5791 —unless a ver y high-impedance, low-capacitance load is being driven—or attenuation is tolerable and predictable. Because the amplifiers are externa l, they can be selected t o optimize for noise, temperature drift, and speed—and the scale factor can be adjusted—depending on the needs of the application. For the reference buffers, the AD8676dual amplifier is recommended, based on its low noise, low offset error, low offset error drift, and low input bias currents. The input bias current specification of the reference buffers is important, as excessive bias currents will degrade the dc linearity. The degradation of integral nonlinear ity, in ppm, as a function of input bias current, is typically: where IBIASis in nA; VREFPand VREFNare in volts. For exa mple, with a ±10-V reference input span, an input bias current of 100 nA will increase the INL by 0.05 ppm. The key requirements for an output buffer are similar to those for the reference buffers—except for bias current, which does not affect the AD5791’s linearity. Offset voltage and input bias current can affect output offset voltage, though. To maintain dc precision, the AD8675is recommended as an output buffer. High-throughput applications require a fast output buffer amplifier with higher slew rate.
Transcript
5/11/2018 20-Bit DAC - slidepdf.com
http://slidepdf.com/reader/full/20-bit-dac 1/4Analog Dialogue 44-04, April (2010)
The 20-Bit DAC Is the EasiestPart of a 1-ppm-AccuratePrecision Voltage SourceBy Maurice Egan
Introduction
A common use o high-resolution digital-to-analog converters(DACs) is providing controllable precision voltages. Applicationsor DACs with resolution up to 20 bits, precision up to 1 ppm,and reasonable speed include gradient coil control in medical MRIsystems; precision dc sources in test and measurement; precisionset-point and position control in mass spectrometry and gaschromatography; and beamprobing in scientic applications.
Over time, the denition o what constitutes a precision integrated-circuit DAC has changed rapidly as semiconductor processingand on-chip calibration technologies have advanced. Once, high-precision 12-bit DACs were considered to be hard to achieve; inrecent years, 16-bit accuracy has become widely available or usein precision medical, instrumentation, and test and measurementapplications; and over the horizon, even greater resolutions andaccuracies are called or in control and instrumentation systems.
High precision applications or integrated circuits now require18-bit and 20-bit, 1-ppm-accurate digital-to-analog converters,a perormance level previously achieved only with cumbersome,expensive, and slow Kelvin-Varley dividers—the preserve o standardslabs and hardly suitable or instrumentation systems in the eld. Moreconvenient semiconductor-based, 1-ppm-accurate solutions to suchrequirements using assemblies o IC DACs have been around orsome years; but these complex systems use many devices, requirerequent calibrations and great care to achieve accuracy, and areboth bulky and costly (see Appendix). The precision instrumentationmarket has long needed a simpler, cost-eective DAC that doesn’trequire calibration or constant monitoring, is easy to use, and oersguaranteed specications. A natural evolution rom 16-bit and 18-bitmonolithic converters—such a DAC is now a reality.
The AD5791 1-ppm DAC
Advances in semiconductor processing, DAC architecturedesign, and ast on-chip calibration techniques make possiblehighly linear, stable, ast-settling digital-to-analog convertersthat deliver better than 1-ppm relative accuracy, 0.05-ppm/ °Ctemperature drit, 0.1-ppm p-p noise, better than 1-ppm long-term stability, and 1-MHz throughput. These small, singlechip devices have guaranteed specications, do not requirecalibration, and are easy to use. A typical unctional diagram orthe AD5791 and its companion reerence- and output buers isshown in Figure 1.
SPIAD5791
1ppm DAC
VREFP
VOUT
VREFN
Figure 1. AD5791 typical operating block diagram.
The AD5791 single-chip, 20-bit, voltage-output digital-to-analogconverter species 1-LSB (least-signifcant bit ) integral nonlinearity (INL)anddierential nonlinearity (DNL), making it the world’s rst monolithic
www.analog.com/analogdialogue
1-ppm-accurate digital-to-analog converter (1 LSB at 20 bione part in 220 = one part in 1,048,576 = 1 ppm). Designeduse in high-precision instrumentation and test and measuremsystems, it oers a signicant leap in all-around perormcompared to other solutions, providing greater levels o accuand repeatability in less space and at lower cost, permitinstrumentation applications that previously would not have beconomically easible.
Its design, shown in Figure 2, eatures precision voltage-mode Rarchitecture, exploits state-o-the-art thin-lm resistor-match
techniques, and employs on-chip calibration routines to ach1-ppm accuracy levels. Because the device is actory calibrated thereore, requires no run-time calibration routines, its latenno greater than 100 ns, so the AD5791 can be used in waveogeneration applications and ast control loops.
VREFPF
V
VREFPS
VREFNF
VREFNS
2R
S0
R
2R 2R
S1
R
2R
S13
R
2R
E62
2R
E61
2R
E0
14-BITR-2R LADDER
6 MSBs DECODED INTO63 EQUAL SEGMENTS
Figure 2. DAC ladder structure.
Besides its impressive linearity, the AD5791 combines 9 nV/noise density, 0.6-μV peak-to-peak noise in the 0.1-Hz to 10requency band, 0.05-ppm/°C temperature drit, and better 0.1-ppm long-term stability over 1000 hours.
A high-voltage device, it operates rom dual supplies o u±16.5 V. The output voltage span is set by the applied positivenegative reerence voltages, V REFP and V REFN , oering a fexchoice o output range.
The precision architecture o the AD5791 requires hperormance external ampliers to buer the reerence sorom the 3.4 kΩDAC resistance and acilitate orce-sensing a
reerence input pins to ensure the AD5791’s 1-ppm linearityoutput buer is required or load driving to unburden the 3.4output impedance o the AD5791—unless a very high-impedalow-capacitance load is being driven—or attenuation is tolerand predictable.
Because the ampliers are external, they can be selected to optimor noise, temperature drit, and speed—and the scale actorbe adjusted—depending on the needs o the application. Forreerence buers, the AD8676 dual amplier is recommenbased on its low noise, low oset error, low oset error dr it,low input bias currents. The input bias current specicatiothe reerence buers is important, as excessive bias currentsdegrade the dc linear ity. The degradation o integral nonlineain ppm, as a unction o input bias current, is typically:
where I BIAS is in nA; V REFP and V REFN are in volts. For examwith a ±10-V reerence input span, an input bias curren100 nA will increase the I NL by 0.05 ppm.
The key requirements or an output buer are similar to thor the reerence buers—except or bias current, which dnot aect the AD5791’s linearity. Oset voltage and input current can aect output oset voltage, though. To maindc precision, the AD8675 is recommended as an output buHigh-throughput applications require a ast output buer ampwith higher slew rate.
http://slidepdf.com/reader/full/20-bit-dac 2/42 Analog Dialogue 44-04, April (2010
Table 1 lists the key speciications o a ew appropriateprecision ampliiers.
The AD5791 oers reduced design time, reduced design risk,reduced cost, reduced board size, increased reliability, andguaranteed specications.
Figure 3 is a circu it schematic implementing the AD5791 (U1)as a precision digitally controlled 1-ppm voltage source with a±10 V range in 20-μV increments using the AD8676 (U2) asreerence buers and the AD8675 (U3) as the output buer.The absolute accuracy is determined by the choice o the
externa l 10 V reerences.
Performance Measures
The important measures o this circuit are integral nonlinearity,dierential nonlinearity, and 0.1-Hz to 10-Hz peak-to-peak noise.Figure 4 shows that typical INL is within ±0.6 LSB.
1.0
–1.0
–0.8
–0.6
–0.4
–0.2
0
0.2
0.4
0.6
0.8
0 200,000 400,000 600,000 800,000 1,000,000
I N L E R R O R
( L S B )
DAC CODE
Figure 4. Integral nonlinearity plot.
Figure 5 shows a typical DNL o ±0.5 LSB; the output isguaranteed monotonic over the entire range o bit transitions.
1.0
–1.0
–0.8
–0.6
–0.4
–0.2
0
0.2
0.4
0.6
0.8
–0.9
–0.7
–0.5
–0.3
–0.1
0.1
0.3
0.5
0.7
0.9
0 200,000 400,000 600,000 800,000 1,000,000
D N L E R R O R
( L S B )
DAC CODE
Figure 5. Differential nonlinearity plot.
Peak-to-peak noise in the 0.1-Hz to 10-Hz bandwidth is about700 nV, as shown in Figure 6.
http://slidepdf.com/reader/full/20-bit-dac 3/4Analog Dialogue 44-04, April (2010)
The AD5791 Is Only the Beginning:
1-ppm Circuit Complexities
Even though precision sub-1-ppm components such as theAD5791 are available on the market, building a 1-ppm systemis not a task that should be taken lightly or rushed into. Errorsources that show up at this level o precision must be careullyconsidered. The major contributors to errors in 1-ppm-accuratecircuits are noise, temperature dri t, thermoelectric voltages, andphysical stress. Precision circuit construction techniques shouldbe ollowed to minimize the coupling and propagation o these
errors throughout the circuit and the introduction o externalintererence. These considerations will be summarized here briefy.Further inormation can be ound in the Reerences.
Noise
When operating at 1-ppm resolutions and accuracies, it is o utmostimportance to keep noise levels to a minimum. The noise spectraldensity o the AD5791 is 9 nV/√Hz, mostly rom the Johnson noiseo the 3.4-kΩ DAC resistance. All peripheral components shouldhave smaller noise contributions to minimize increases to the systemnoise level. Resistor values should be less than the DAC resistanceto ensure that their Johnson noise contribution will not signicantlyadd to the root-sum-square overall noise level. The AD8676reerence buers and the AD8675 output buer have a speciednoise density o 2.8 nV/√Hz, well below the DAC’s contribution.
High-requency noise can be eliminated relatively easily withsimple R-C lters, but low-requency 1/ noise in the 0.1-Hz to10-Hz range cannot be easily ltered without aecting dc accuracy.The most eective method o minimizing 1/ noise is to ensurethat it is never introduced into the circuit. The AD5791 generatesabout 0.6 μV p-p o noise in the 0.1-Hz to 10-Hz bandwidth, wellbelow the 1-LSB level (1 LSB = 19 μV or a ±10-V output span).The target or max imum 1/ noise in the entire circuit shouldbe about 0.1 LSB, or 2 μV; this can be ensured through propercomponent choice. The ampliiers in the circuit generate0.1-μV p-p 1/ noise; the three ampliers in the signal chaingenerate a total o approximately 0.2-μV p-p noise at the circuitoutput. Add this to the 0.6-μV p-p rom the AD5791, and the
total expected 1/ noise is about 0.8 μV p-p, a gure that closelycorrelates with the measurement displayed in Figure 5. This oersadequate margin or other circuitry that may be added, such asampliers, resistors, and a voltage reerence.
Besides random noise, it is important to avoid errors caused byradiated, conducted, and induced electrical intererence. Suchtechniques as shielding, guarding, and scrupulous attention togrounding and proper printed-circuit-board wiring techniquesare imperative.
Temperature Drift
As with all precision circuits, drit o all components withtemperature is a major source o error. The key to minimizingthe dri t as much as possible is to choose critical components with
sub-1-ppm temperature coecients. The AD5791 exhibits a verylow 0.05-ppm/°C temperature coecient. The AD8676 reerencebuers drit at 0.6 μV/°C, introducing an overall 0.03-ppm/ °Cgain dri t into the circuit; the AD8675 output buer contributesa urther 0.03-ppm/°C output drit; this all adds up to a gureo 0.11 ppm/°C. Low drit, thermally matched resistor networksshould be used or scaling and gain circuits. Vishay bulk metal-oil voltage-divider resistors, series 300144Z and 300145Z, witha temperature coecient o resistance tracking to 0.1 ppm/ °C,are recommended.
Thermoelectric Voltages
Thermoelectric voltages are the result o the Seebeck eect:temperature-dependent voltages are generated at dissimilar
metal junctions. Depending on the metallic components ojunction, the generated voltage can be anywhere rom 0.2 μVto 1 mV/°C. The best case, a copper-to-copper junction, generate less than 0.2 μV/°C o thermoelectric EMF. In the wcase, copper-to-copper-oxide can generate up to 1 mV/°Cthermoelectric voltage. This sensitivity to even small temperafuctuations means that nearby dissipative elements or slow-moair currents crossing over a printed circuit board (PCB) can crvarying temperature gradients, which in turn generate varthermoelectric voltages that are maniested as a low-reque
drit similar to low-requency 1/ noise. Thermoelectric voltcan be avoided by ensuring that there are no dissimilar junctin the system and/or eliminating thermal gradients. While virtually impossible to eliminate dissimilar metal junctions—mdierent metals exist in IC packaging, PCB circuits, wiring,connectors—keeping all connections clean and oxide-ree wia long way to keeping thermoelectric voltages low. Enclosingcircuit to shield circuitry rom air currents would be an eecthermoelectric voltage stabilizing method, and it could haveadded value o providing electrical shielding. Figure 7 showsdierence in voltage drits between a circuit that is open tocurrents and one that is enclosed.
1.0
–1.0
–0.8
–0.6
–0.4
–0.2
0
0.2
0.4
0.6
0.8
0 35030025020015010050
V O L T A G E
( V )
TIME (Seconds)
CIRCUIT IN FREE AIR
CIRCUIT ENCLOSED
Figure 7. Voltage drift vs. time for open- and enclosed syste
To cancel out the thermoelectric voltages, compensating junctcould be introduced into the circuit, a task that would invconsiderable trial and error and iterative testing to ensurecorrect pairing and positions o the inserted junctions. By armost ecient method is to reduce the number o junctions incircuit by minimizing component count in the signal path stabilizing the local and ambient temperatures.
Physical Stress
High-precision analog semiconductor devices are sensitivstress on their package. Stress relie compounds used withinpackaging have a settling eect, but they cannot compensat
signicant stress due to pressure exerted directly on the packby local sources, such as fexing o the PCB. The larger the princircuit board, the more stress that a package could potentsuer, so sensitive circuitry should be placed on as small a boas possible—with connection to the larger system through fexor nonrigid connectors. I a large board cannot be avoided, srelie cuts should be made around sensitive components, onor (preerably) three sides o the component, greatly reducingstress on the component due to board fexing.
Long-Term Stability
Ater noise and temperature drit, long-term stability deseconsideration. Precision analog ICs are very stable devices buundergo long-term age-related changes. Long-term stability o
5/11/2018 20-Bit DAC - slidepdf.com
http://slidepdf.com/reader/full/20-bit-dac 4/44 Analog Dialogue 44-04, April (2010
AD5791 is typically better than 0.1 ppm/1000 hours at 125°C. Theaging is not cumulative but ollows a square root rule (i a device agesat 1 ppm/1000 hours, it ages at √2 ppm/2000 hours, √3 ppm/3000hours, …), and the time is typically 10 times longer or each 25°Creduction in temperature; so, at 85°C operation, one can expectaging o 0.1 ppm over a 10,000 hour period, approximately 60 weeks.I this is extrapolated, 0.32 ppm aging can be expected over a 10-yearperiod, so the data sheet dc specications can be expected to dritby 0.32 ppm over a 10-year period when operating at 85°C.
Circuit Construction and LayoutIn a circuit where such a high level o accuracy is important,careul consideration o the power supply and ground returnlayout helps to ensure the rated perormance. Design the PCB suchthat the analog and digital sections are separated and connedto separate areas o the board. I the DAC is in a system wheremultiple devices require an analog-to-digital ground connection,establish the connection at one point only. Establish the star-pointground as close as possible to the device. There should be amplepower supply bypassing o 10 μF in parallel with 0.1 μF on eachsupply terminal, as close to the package as possible, ideally rightup against the device. The 10-μF capacitors should be o thetantalum bead type. The 0.1-μF capacitor should have low eect iveseries resistance (ESR) and low eective series inductance (ESL),
such as the common multilayer ceramic types—to provide a low-impedance path to ground at high requencies to handle transientcurrents due to internal logic switching. A series errite bead oneach power supply line will urther help to block high-requencynoise rom getting through to the device.
The power supply traces should be as large as possible to providelow-impedance paths and reduce the eects o glitches on thepower-supply line. Shield ast-switching signals, such as clocks,with digital ground to avoid radiating noise to other parts o theboard. They should never be run near the reerence inputs or underthe package. It is essential to minimize noise on the reerenceinputs because it couples right through to the DAC output. Avoidcrossover o digital and analog signals, and run traces on oppositesides o the board at right angles to each other to reduce the eectso eedthrough on the board.
Voltage Reference
Holding the perormance o the entire circuit rmly within itsgrasp is the external voltage reerence; its noise and temperaturecoecient directly impact the system’s absolute accuracy. Tocapitalize on the challenge posed by the 1-ppm AD5791 digital-to-analog converter, the reerence and assoc iated componentsshould have temperature drit and noise speciicationscomparable to those o the DAC. Although a reerence withtemperature drit o 0.05 ppm/°C is nothing short o antasy,1 ppm/°C and 2 ppm/°C voltage reerences with 0.1-Hz to 10-Hznoise o less than 1 μV p-p do exist.
ConclusionAs the accuracy requirements o precision instrumentation—andtest and measurement applications—increase, more accuratecomponents are being developed to meet these needs. They haveguaranteed precision specications at the 1-ppm level withouturther user calibration and are easy to use. However, whendesigning circuitry or this level o precision, one must bear inmind the many environmental and design-related challengesthat exist. Successul precision-circuit perormance will come asa result o considering and understanding these challenges andmaking correct component choices.
References
(Inormation on all ADI components can be ound at www.analog.com.)
1. “The Long Term Stability o Precision Analog ICs, or How toAge Graceully and Avoid Sudden Death.” Analog DevicesRarely Asked Questions.
pd/LowLevMsHandbk_1.pd .3. MT-031, Grounding Data Converters and Solving the Mystery o
“ AGND” and “DGND.” http://www.analog.com/static/imported-les/tutorials/MT-031.pd .
Author
Maurice Egan [[email protected]] is anapplications engineer with the Precision ConvertersProduct Technology Group based in Limerick.Maurice joined Analog Devices in 1998 and holds aBEng in electronic engineering rom the Universityo Limerick, Ireland.
Appendix Figure 8 shows a block diagram o a typical contemporary1 ppm DAC solution. The core o the circuit consists o two16-bit digital-to-analog converters—a major DAC and aminor DAC—the outputs o which are scaled and combinedto yield an increased resolution. The major DAC output issummed with the attenuated minor DAC output so that theminor DAC lls the resolution gaps between the major DAC’sLSB steps.
16-BIT
MINOR DAC
16-BIT
MAJOR DAC
E R R O
R C O R R E C T I N G
S O F T
W A R E E N G I N E
ATT
24-BIT
ADC
VOUT
Figure 8.
The combined DAC outputs need to be monotonic, but notextremely linear, because high perormance is achieved withconstant voltage eedback via a precision analog-to-digitalconverter, which corrects or the inherent component errors;thus, the circuit accuracy is limited by the ADC rather thanthe DACs. However, because o the requirement or constantvoltage eedback and the inevitable loop delay, the solution isslow, potentially requir ing seconds to settle.
Although this circuit can, with signicant endeavor, ultimatelyachieve 1 ppm accuracy, it is complex to design, likely torequire multiple design iterations, and requires a sotwareengine and precision ADC to achieve accuracy. To guarantee1-ppm accuracy the ADC will also require correction—sincean ADC with guaranteed 1-ppm linearity is not available. Thesimple block diagram o Figure 8 illustrates the concept, butthe actual circuit in reality is ar more complex, with multiplegain, attenuation, and summing stages, involving manycomponents. Also required is signicant digital circuitry toacilitate the interace between both DACs and the ADC—notto mention the sotware required or error correction.
