Design femtoampere circuits with low leakage, part one
Carefully apply materials science while using guards and shields to reduce leakage.
Paul Grohe, Texas Instruments -- EDN, November 17, 2011
At A Glance
|
Circuits that carry femtoamperes of current have
many subtleties that you wouldn’t normally consider
in the design and layout of conventional circuitry.
If you overlook these subtleties, the circuit loses
low-end resolution and exhibits drift due to the
components, materials, and circuit layout. Knowing
the circuit’s limitations and leakages and providing ways to minimize
or eliminate them will lead to improved circuit performance.
The world below a picoampere is unique and plays by a different set of rules. In this world, even the mechanical parts of the circuit can become parts of the electrical circuit. Designing for operation at subpicoamp and femtoamp levels requires special techniques and compromises that normal current levels don’t generally require (Reference 1). Unfamiliarity with or neglecting these precautions can result in endless headaches for designers. Electrical engineers will find themselves playing double roles as mechanical engineers.
This three-part article guides you through the tricky and unconventional design techniques you need to create successful low-current circuits. This first part defines and describes the designs that carry these low currents. It explains the problems that arise when you design these circuits and examines the application of shielding and guarding methods. Part two will examine how your component selection affects the performance of your low-leakage circuits and discuss how noise creeps into low-leakage designs. Part three will provide detailed PCB-design techniques and show a realworld example of a low-leakage design. It will also describe how to verify the performance of your low-leakage-design techniques.
Low-current applications
To put things into perspective, 1A equals 6,241,500,000,000,000,000, or 6.2418 electrons/sec; 1 pA, or 1−12A, equals 6.24 million electrons/sec; and 1 fA equals 1−15A, or 6240 electrons/sec. In the subpicoamp world, there are three common enemies: current leakages, noise sources, and stray capacitance. A good low-current design must minimize the effects of these common enemies and strike a balance between optimal performance and product manufacturability. You will need special techniques and materials that may be incompatible with conventional production flows.
These high-impedance circuits
often go directly into an amplifier input
with no parallel-resistive connections.
Examples of these circuits include pH
probes, gas-sensor amplifiers, medical
sensors, sample-and-hold circuits,
and three-amplifier instrumentation
amplifiers. The circuits can have input
impedances into the teraohm range. A
transimpedance amplifier, or current-to-voltage converter, is often used at
these low current levels. You see this
circuit configuration in noninverting
amplifiers, photodetector amplifiers,
current-to-voltage converters, and photomultiplier
circuits. The amplifier’s
inverting input node and its feedback
elements are critical nodes. The current
leakage in this node determines
the ultimate accuracy of the device.
Higher-current circuits, such as low-frequency
filters and logarithmic
amplifiers, also benefit from low-leakage-design techniques. They will
have extended dynamic range, with
improved low-end accuracy and lower
drift than nonoptimized designs.Causes of disturbance
Dirty PCB traces can cause leakage at low currents. The dirt between the traces or across insulating materials—not the trace or wire itself—causes the leakage, serving as a conductive medium between two conductors. Dry dirt in itself may not cause a problem. A combination of dirt with moisture, salts, and oil, however, becomes conductive. The concept here is simple: Keep things clean.
Moisture is the instigator of most leakage problems. When moisture combines with environmental salts and other contaminants, its conductivity increases. Insulation, PCBs, and other hydroscopic materials absorb the moisture, decreasing the electrical resistance of the materials and leading to increased leakage between the conductors.
Contamination between conductors can also create a galvanic reaction in the presence of the right combination of materials and moisture. Moist and salty dirt between a copper trace and a zinc-plated screw or an aluminum case will generate a current between the materials. This current is detrimental to your measurement and causes corrosion of the materials. Because the moisture level varies over the course of the day, season, and geographical location, it creates a moving baseline leakage that is difficult to remove. These leakages change hourly, weekly, or yearly, depending on the environment and the season.
The particles and moisture in air as it moves over a conductor generate a small charge, so you should protect the input circuit from moving air currents. Make sure that fan-cooling airflow does not blow directly over sensitive nodes. Airflow can also cause dust and moisture to accumulate on the conductors and components.
You must take into account the properties of insulating materials in your design. These materials come into direct contact with low-level signals, usually through the connectors, the supports, or the PCB. In the electronics industry, the most common insulators are fiberglass, glass, ceramic, PVC (polyvinyl chloride), epoxy, and Teflon. Each material has its own weaknesses and strengths. Dry air is a good insulator. Keeping conductors in the air can provide the lowest-leakage results. Air does have a low breakdown voltage, however, which limits this technique in high-voltage applications. PTFE (polytetrafluoroethylene) and FEP (fluorinated-ethylene propylene), more commonly known as Teflon, have the best leakage and high-voltage characteristics of common insulating materials, but they are expensive, soft, and difficult to machine. Teflon PCBs are expensive because of the material and the extra steps the fabrication process requires.
Ceramic, although a good insulator, tends to be piezoelectric. Ceramic self-generates charge when it is subjected to stress or impact. It also readily absorbs moisture if it is not sealed or glazed. Although glass is a good insulator, it displays some of the piezoelectric properties of ceramics. IC packages use a molded glass-epoxy compound that allows for currents lower than 1 fA. Epoxy is an excellent, low-cost insulator; however, it is hydroscopic and can absorb moisture over time. Many components, connectors, and wire insulation use PVC, which can generate charge if flexed or rubbed against another conductor, just as combing your hair can generate current. For this reason, PVC insulation in and around the input circuit should be avoided.
It might seem logical to build the ultimate low-leakage layout entirely on a slab of Teflon. This can be a bad idea, however. Because Teflon is a good insulator, any charge deposited on its surface will slowly dissipate. If a sensitive node is nearby, the accumulated charge will lead to slow settling or drift. A better approach is to cover a large surface area with a guarded conductive plane. Although this approach seems counterintuitive to the desire for low leakage, you should minimize the use of insulators. Insulation must provide isolation, but using too much of it provides a surface to accumulate extra charge.
For low-voltage circuits, an aluminum standoff topped with a small piece of Teflon insulation works better and is less expensive than using an entire standoff made of Teflon. If the circuit will be handling high voltages, you need a Teflon standoff for its better insulating properties. For ac circuits, the narrow insulator has higher stray capacitance that may cause other problems. As in all analog design, you must consider many trade-offs. PCBs have a large influence on low-leakage design because the PCB material is in intimate contact with all of the circuit nodes. The performance of your circuit is only as good as the performance of the PCB material. As with RF circuits that operate at gigahertz speeds, you should consider the PCB as an active component. Most PCBs’ material characteristics and development focus on high-frequency RF designs. Manufacturers gear PCB specifications toward circuits that operate at these speeds. They give a nod to low-current requirements by specifying a volume resistivity. The manufacturer’s specifications are for the fresh laminate material before processing—not the finished product. That product comprises a sandwich of laminates, bonding glues, fillers, solder masks, and silk-screen that make up a PCB.
The most common PCB material is FR4 (flame-resistant 4), which comprises epoxy-impregnated fiberglass cloth. Manufacturers compress this epoxy under high pressure to form a solid board. FR4 has good electrical properties, but it is not the most desirable material for low-current circuits. You can improve the performance of FR4 using special layout and circuit techniques.
When performance is more important than cost, you can use exotic Teflon or ceramic hybrids, such as Rogers Corp’s Duroid hybrid substrate materials, targeting use in microwave and ultra-high-speed digital circuits. The materials’ excellent controlled dielectric properties can result in two- to three-times-lower stray capacitance and leakage than those of FR4, but at a cost two- to five-times higher.
The boards also require special PCB-fabrication processes and etching, which some PCB-fabrication houses may be unable to accommodate. The Rogers soft, bendable 3003 material, which is ceramic-reinforced PTFE, requires backing for mechanical stability. Rogers 5880, a glass-reinforced PTFE, gives the best low current and stray capacitance, but it is brittle and cracks easily. It is possible to create a hybrid board, with advanced materials for the critical layers and FR4 for noncritical layers and mechanical stability. This approach is expensive and requires using an advanced board house, however.
Use caution with solder-mask placement. Although solder masks generally help reduce moisture infiltration into the PCB material, surface-charge problems might arise with large areas of Teflon. A better approach uses a bare-copper guard-plane area around sensitive nodes. To prevent oxidation, either solder-level or plate the bare-copper guard area with gold or tin.
Shield, guard, encloseYou wire a metallic shield, case, or enclosure
to a ground or a common potential.
At high impedances, however,
these shields create problems with stray
capacitance and leakages. Examine, for
example, a circuit with a 2.5V input voltage
and with 2.5V across the stray-capacitance
and leakage paths (Figure 1). The
2.5V across the leakage resistance creates
a leakage current, and the 2.5V source
voltage charges or discharges the stray
capacitance, which takes some time to get
through the high source impedance and
affects the measurement’s settling time.
Guards are important in subpicoamp
designs because they can cancel
the input-leakage currents and most
of the fixture capacitance. You drive the
guard to a potential equal to the input-signal
level. You apply a buffered output
derived from the measurement amplifier.
This guard acts as a subshield, surrounding
and protecting the
input-signal lines. External
leakages now flow into the
low-impedance guard instead
of the input traces (Figure
2). This approach yields only
a few millivolts of potential
difference instead of 2.5V
across and smaller current
flows through leakage resistor
RLEAK and stray-capacitance
capacitor CSTRAY. As a bonus,
guards also reduce the input
capacitance through a bootstrapping
effect. Performed
correctly, this approach can
cancel out fixture and cable
capacitance. Unfortunately,
you cannot cancel out
the amplifier’s input-stage
capacitance.Locate the input traces
and all of the sensitive feedback
components on your
PCB within the perimeter of
the thick copper-guard traces
(Figure 3). Then, remove the
solder mask from this area to
reduce surface charges. Buffer
amplifier A2 drives the guard ring. In the inverting and transimpedance
designs, you drive the guard to
the same potential as the noninverting
input’s node and feed the potential on the
noninverting pin to the guard buffer. The
noninverting node is low-impedance,
and the buffer does not affect the circuit’s
operation. The guard should cover the
entire input section, the inverting node,
and the feedback resistor. Extend it as far
into the sensor circuit as possible without
affecting sensor operation.
When designing in the noninverting
mode, drive the guard to the same
potential as the inverting input node
through a buffer. This node follows
the input signal through the feedback
action of the amplifier. Take care that
the capacitance of the buffer’s input
does not cause peaking due to capacitive
loading of the inverting node.
The guard-driver amplifier should be
unity-gain-capable and protected from
short circuits and external overvoltages.
The bandwidth of the buffer should be
slightly wider than the main circuit’s
bandwidth to reduce phase-lag errors.
Avoid a peaking response in the guard
buffer to prevent system instability. A
grounded shield protects the circuit
from external noise and EMI by shunting
the noise to ground. Because the
grounded shield generally does not follow
the input voltage, it does not cancel
the capacitance caused by the guard.In the previous examples, you buffer the guard line from a circuit node using a separate amplifier, providing a low impedance to drive the shields and coaxial-cable guards. If you need to guard a small location, you can derive a local guard from the opposite input terminal. Keep in mind that the local guard also adds capacitance to the node to which it connects. This capacitance can lead to peaking in noninverting-amplifier configurations. If the opposite node is high-impedance, the guard can introduce external noise into the summing node unless you shield the guard itself. Do not use an unbuffered guard to drive external circuitry. Use it only for the immediate area surrounding the device.
Keep in mind that the guard is not ground, and ground is generally not a guard. The guard lines should not carry any currents other than the leakages, and you should treat them as signal lines. For effective designs, use guards and grounds together. The guard surrounds the input trace, and the grounded shields protect the guards from external interference. When you lay out a PCB, place a guard plane or guard traces below the sensitive traces. Be careful not to break up the power or ground layer too much. Surround the input circuit in a guarded cocoon using metallic shields on visible component sides and guard traces on layers below the sensitive nodes.
You should enclose your low-current circuits in a sealed environment. If possible, include a desiccant pack to absorb any traces of moisture. The wiring and control shaft entry and exit points should be airtight. You can use triaxial cables and connectors for low-current measurements. The cable contains both an outer grounding shield and an inner guard shield around the center conductor, extending the guard out to the measurement point. Commercial test equipment often uses Trompeter 70-series triaxial BNCs. Agilent prefers the three-lug style, while Keithley prefers the two-lug style.
|
Reference |
|
Author’s biography
Paul Grohe is an applications engineer for Texas Instruments’ Precision Systems group. He attended the College of San Mateo (San Mateo, CA).
|
For More Information |
|||
|
Agilent Technologies |
Keithley Instruments |
Rogers Corp |
Trompeter |
Talkback
-
Guard Rings, Transmission Lines and Space Cloth
Paul, your article reminded me that those of us who work with guard rings probably know more than we think we know about quantitative performance of guard rings. It turns out that computing the leakage resistance due to surface resistivity is mathematically the same problem as calculating the characteristic impedance of a transmission line made up of long parallel ideal conductors in vacuum. You can compute the leakage from the guard ring to node it is guarding if the geometry is the same as a transmission line that has a known formula for its characteristic impedance. If you have a vacuum dielectric transmission line made of ideal long straight parallel conductors, you can perfectly terminate it by butting it into space cloth at a right angle. In the case where one conductor surrounds the other, like a coax, you can cover the end with space cloth and the transmission line will be perfectly terminated. What is space cloth? It is a hypothetical flat surface with a resistivity of 377 Ω / square. The impedance of a vacuum coax is 138 * log ( C/c ) with upper case C being the inside circumference of the outer conductor and lower case c being the outer circumference of the inner conductor. Of course C/c = R/r = D/d where the R's are radii and the D's are diameters.
So, suppose the node you wish to guard has a circumference of c and the inner edge of the guard ring has a circumference C. Compute the characteristic impedance of the transmission line of the same geometry. Let's say it is 80 ohms. Now, the surface resistivity of your board isn't 377 Ω / square. It is a lot more. Let's say your board has a surface resistivity of 29 MΩ / square. So then the actual resistance between the guard ring and the node it protects is 80 x (29,000,000 / 377) or about 6 MΩ. Now let's suppose the guard ring is at ground potential and the guarded node is at 1 mV. Then we would have about 167 pA current into the guarded node.
But we can do more. Let's ask what would happen if there were no guard ring. Let's suppose there was another node at 5 volts having the same diameter as the node that needs a guard. Further let's suppose the center to center separation of the two nodes was 4 diameters. We look up the formula for the impedance of a transmission line composed to two round wires. It is about 250 ohms. Using the same sort of ratio calculation we get that the actual resistance between the nodes is 250 x (29,000,000 / 377) ≈ 20MΩ. With a 5 volt difference, the current flow to the guarded node would be 250 nA. So, the guard reduced the leakage by a factor of 1500. This improvement factor holds no matter what the actual surface resistivity is.
Roy McCammon - 2011-28-11 21:53:05 PST -
PS: I failed to mention that Landis et al. had removed the JFET from its conventional packaging, with the device still sitting on top of its gate lead, and presumably used some lower-loss dielectric as a gate lead support. Thus the lead was no longer surrounded by the borosilicate glass. The paper was in IEEE Transactions on Nuclear Science, 1971, NS-18, pp. 115-124. It appears another paper about the light-reset technique was written with Goulding as the lead author in 1969; this may not have been as successful (perhaps the JFET was left in the package and just the top removed?) as I don't remember seeing it in those days.
Brad Wood - 2011-18-11 08:27:15 PST -
In the early days of Teflon, and possibly to this day, it was noticed that when subjected to mechanical stress the material could be slightly piezoelectric and produce currents in the low-pA range that persisted for minutes. So even this wonderful insulator is not without potential problems.
Quite a few years ago, when I was working on cooled JFET charge amplifiers for UCLA, I devised a pulsed-reset system which used a small-geometry RF bipolar as a switched current pulse generator. I remarked to a colleague of a particular professor, who was a vigorous self-promoter that regarded himself as an expert on infrared instrumentation, that the parts had collector-base leakage of less than 100fA at room temperature, decreasing markedly with cooling as anticipated. He said "Femtoamperes!! In this business you have to be concerned with tiny currents --- picoamperes!" He was actually unfamiliar with those units, and I realized that the proverbial emperor was at best scantily clad. When the professor insisted that anything I did be vetted by him, lest a mistake occur of which the prof. would get kidded mercilessly about, I began to lay my plans to leave the department, which I managed several months later.
Another detail not covered in this excellent note is lossy dielectic noise, per se. This is important for charge amplifiers when other noise sources like FET gate leakage and noise charge associated with channel thermal noise, and detector leakage noise, have been reduced by careful design and cooling. Years ago a group at Lawrence Berkeley published on a photonically-reset preamp and noted a large improvement in noise charge, which they attributed to the elimination of the feedback resistor. Then Kern and Mackenzie at Bell Labs pointed out that the attribution was improbable, and published results for their own preamp in which the JFET was repackaged in a low-loss material (BeO, Al2O3, or BN) and achieved similar results :) Bell Labs had a nasty habit of doing that sort of thing in those days.
Brad Wood - 2011-17-11 09:46:46 PST






















