100nF

PCB leakage calculator: guard ring and source error

5 V across 1 GΩ of board is 5.00 nA, and 5.00 nA into a 100 MΩ sensor is 500 mV of error. This calculator adds the board path, a switch's on-leakage and the amplifier's bias current at a high-impedance node, turns them into voltage through the source and switch resistance by TI SCDA042's Equation 1, and shows what a guard ring driven by a buffer from SCDA042's Table 2-1 leaves — then the largest source resistance that keeps the error inside your budget.

budget 25.0 µVunguardedguarded100 V1 nV1 kΩ10 GΩsource resistance
Fig 1 — at 100 MΩ the board leaves 500 mV of error unguarded and 166 µV guarded, against a 25.0 µV budget.
Board leakage, unguarded
5.00 nA (5 V across 1.00 GΩ)
Error at the node, unguarded
500 mV from 5.00 nA through 100 MΩ
Board leakage, guarded
60 fA (60.0 µV left by OPA397)
Error at the node, guarded
166 µV from 1.66 pA
What the guard removes
83,333× less board current
Share of the node current from the board
100.0 % unguarded, 3.6 % guarded
Largest source within 25 µV
5.00 kΩ unguarded, 15.1 MΩ guarded

Most of what remains is the parts' own leakage, which no guard removes: a lower-leakage switch or amplifier, a lower source resistance, or a cooler board. SCDA042: "the application needs to start with a very high precision capable device".

How this is calculated

Standard: TI SCDA042, Guarding in Multiplexer Applications (May 2022): p. 2, Equation 1 and Tables 2-1, 3-1, 3-2, 4-1; TI SBOA590, Op Amp Offset Voltage and Bias Current Limitations (Nov 2024): Tables 1-1 and 2-1, pp. 18 and 33–34

Iboard=ΔVRleakI_{board} = \frac{\Delta V}{R_{leak}}
Ohm's law across the board path, as SCDA042 p. 2 applies it: "a 5 V potential difference between two traces ... an estimated trace to trace resistance in the Gigaohm range, this would amount to nA's worth of leakage."
Iboard, guarded=VOS, guardRleakI_{board,\,guarded} = \frac{V_{OS,\,guard}}{R_{leak}}
Derived. A driven guard holds the copper round the node "at the same potential" (SCDA042 p. 2); what it misses by is the guard buffer's offset, the figure SCDA042 Table 2-1 lists its buffers by: OPA397 60 µV, OPAx197 25 µV, OPA593 10 µV, OPA455 3.4 mV.
VERROR=(RON+RSOURCE)⋅ILEAK,ILEAK=Iboard+ION+IB  (+ IB, guard)V_{ERROR} = \left(R_{ON} + R_{SOURCE}\right) \cdot I_{LEAK}, \qquad I_{LEAK} = I_{board} + I_{ON} + I_{B} \;\left(+\, I_{B,\,guard}\right)
SCDA042 Equation 1 (p. 4). I_ON is a switch's on-leakage (Tables 3-1 and 4-1), I_B the amplifier's bias current; with a guard, the guard buffer's own bias current is added, since its input sits on the node (Figure 2-4). A guard changes only I_board.
RSOURCE, max=VbudgetILEAK−RONR_{SOURCE,\,max} = \frac{V_{budget}}{I_{LEAK}} - R_{ON}
Equation 1 solved for the source resistance at which the error reaches the budget. Negative when R_ON × I_LEAK alone exceeds it.
Ileak(T)=Ileak(25 ∘C)⋅2(T−25 ∘C)/10 ∘CI_{leak}(T) = I_{leak}(25\,^{\circ}\mathrm{C}) \cdot 2^{(T - 25\,^{\circ}\mathrm{C})/10\,^{\circ}\mathrm{C}}
SBOA590 p. 18: "the leakage of a silicon diode doubles every 10°C". Used only in the comparison below the calculator; the switch list takes each part's hot figure from SCDA042 Table 3-1 instead.

Assumptions

What it computes

A high-impedance node is surrounded by copper at other potentials, and the board between them is a resistor nobody placed. TI's guarding note, SCDA042, lists what forms it — flux, debris, "the inherent board resistance and dielectrics between conductors" — and works the arithmetic: "At just a 5 V potential difference between two traces on a PCB and using an estimated trace to trace resistance in the Gigaohm range, this would amount to nA's worth of leakage". At 1 GΩ that is 5.00 nA.

Current is not yet error. It becomes error by flowing through the resistance at the node, and SCDA042 gives the conversion as its Equation 1, VERROR = (RON + RSOURCE) × ILEAK: the switch's on-resistance plus the source behind it. The same 5.00 nA is 250 nVon a 50 Ω node and 500 mV on the 100 MΩ of a pH sensor, the figure TI's offset paper SBOA590 gives for one (p. 34). The calculator adds every current into the node — the board path, a switch's on-leakage, the amplifier's bias current — and puts the total through Equation 1.

A guard attacks the voltage, not the resistance. Copper round the node, driven by a buffer to the node's own potential, leaves almost nothing across the board path: SCDA042 says guard and net at the same potential create "an incredibly small voltage delta". How small is the buffer's offset, which is why SCDA042's Table 2-1 lists its recommended guard buffers by offset voltage. With the OPA397's 60.0 µVin place of 5 V, the board current falls by 83,333 times, to 60 fA.

What a guard cannot touch is the leakage the parts make themselves. A switch's on-leakage flows inside the package, and the amplifier's bias current flows into its own pin; the guard only stops current crossing the board. The buffer that drives the guard adds its own bias current to the node, which is one reason SCDA042 asks for "a precision buffer to reduce any extra leakage currents emanating from the op-amp itself".

Worked example: a pH probe beside a 5 V rail

The calculator's defaults. A 100 MΩ source sits 5 V from a neighbouring trace across 1 GΩ of board, read by an amplifier with the OPA392's 0.8 pA maximum bias current (SBOA590 Table 2-1, at 25 °C), against the OPA192's 25 µV offset as the budget (SBOA590 Table 1-1):

Board path, unguarded   5 V / 1 GΩ                          = 5.00 nA
Amplifier bias          OPA392, 25 °C max                   = 800 fA
Error, unguarded        100 MΩ × 5.00 nA                 = 500 mV
Board path, guarded     60.0 µV (OPA397) / 1 GΩ            = 60 fA
Node current, guarded   60 fA + 0.8 pA + 0.8 pA (buffer)      = 1.66 pA
Error, guarded          100 MΩ × 1.66 pA                  = 166 µV
Largest source, guarded 25 µV / 1.66 pA                   = 15.1 MΩ

The guard takes the error from 500 mV to 166 µV, and still the budget fails. Guarded, the board is 3.6 % of what is left; the rest is the two bias currents. Past this point the layout has done its part, and the remaining fixes are a lower-bias amplifier or a lower source resistance — the guarded node can hold 25 µV only up to 15.1 MΩ. Without the guard the same budget allows 5.00 kΩ: 25 µV ÷5.00 nA, the crossover the guard-ring article plots.

Error by source resistance and guard buffer

The board path alone, 1 GΩ with 5 V on the far side, through Equation 1 with no switch. Each guard row replaces the 5 V with that buffer's offset from SCDA042 Table 2-1. The 100 MΩ cell of the first row is the worked example's unguarded error, less its 0.8 pA of bias current.

Board path1 kΩ10 kΩ100 kΩ1 MΩ10 MΩ100 MΩ1 GΩ
No guard5.00 µV50.0 µV500 µV5.00 mV50.0 mV500 mV5.00 V
OPA397 guard, 60 µV60.0 pV600 pV6.00 nV60.0 nV600 nV6.00 µV60.0 µV
OPAx197 guard, 25 µV25.0 pV250 pV2.50 nV25.0 nV250 nV2.50 µV25.0 µV
OPA593 guard, 10 µV10.0 pV100 pV1.00 nV10.0 nV100 nV1.00 µV10.0 µV
OPA455 guard, 3.4 mV3.40 nV34.0 nV340 nV3.40 µV34.0 µV340 µV3.40 mV

Read across a row and the error rises tenfold per decade of source resistance; read down a column and each buffer's offset sets its own floor. The OPA455, the 150 V part in the table, guards with 3.4 mV of offset and leaves 57 times the OPA397's current: a high-voltage guard buffer is chosen for its range, and its offset is the price.

SCDA042's multiplexer table, reproduced

Table 3-2 of the note converts its two multiplexers' leakage into offset error "assuming Rsource is negligible", which is Equation 1 with RON alone. The calculator gives:

Part25 °C125 °CTable 3-2 prints
TMUX11087.50 pV4.66 nV7.5 pV / 4.6 nV
TMUX130875.0 nV216 µV75 nV / 216 µV

One cell differs in the last digit: 950 pA × 4.9 Ω is 4.655 nV, which the table prints as 4.6 nV, cut rather than rounded. The calculator keeps the product. The point of the table survives either way: the TMUX1308's 216 µV at 125 °C comes from RON alone, before any source resistance multiplies it.

Leakage at temperature: the rule and the datasheet

SBOA590 gives the rule of thumb behind hot leakage: "In general, the leakage of a silicon diode doubles every 10°C", which over 25 to 125 °C is a factor of 1024, and it puts CMOS input bias current up "by a factor of 1000× between room temperature and 125°C". SCDA042's own table checks it:

PartOn-leakage, 25 °C125 °C, Table 3-1RatioDoubling rule, 125 °C
TMUX11083.00 pA950 pA317×3.07 nA
TMUX13081.00 nA800 nA800×1.02 µA

The rule overstates both, by 3.2 times for the TMUX1108. It is the right shape — two to three orders of magnitude over a hundred degrees — and the wrong number for any particular part. Where the datasheet gives a hot figure, the calculator's switch list uses it, which is why the two SCDA042 multiplexers appear twice.

Where the model stops

Further reading