ADC Input Buffer Circuit Design

A successive approximation converter samples the input onto a small capacitor at the start of every conversion. The buffer in front of it exists to supply the charge that the sampling takes, and to present a defined impedance to the source.

Why a Buffer Is Needed

At the sampling instant the converter connects its internal capacitor to the input pin. The charge that flows comes from whatever is connected, and if the source impedance is high the voltage at the pin dips and the conversion is wrong.

The dip is not visible on a slow instrument and it is present on every conversion. It appears as a gain error, a non linearity or a dependence on the previous channel in a multiplexed system.

A buffer with a low output impedance and enough bandwidth to recover from the charge step before the sampling window closes removes the problem. The alternative is to slow the converter down until the source can settle, which is not always possible.

settling time and the Sampling Window

settling time is the time the buffer takes to recover to within a fraction of a least significant bit after the charge step. It is the sum of the slew limited part and the exponential part, and the second dominates for a small step.

The available time is the sampling window of the converter, which is quoted in the datasheet and can be a few tens of nanoseconds at a high throughput. The buffer has to settle within that time into the load it actually sees.

A buffer with a bandwidth of a few megahertz settles to a sixteenth of a bit in a microsecond, which is comfortable. The same buffer driving a multiplexer with a large stray capacitance settles far more slowly.

charge injection and Kickback

The sampling switch also injects a small charge of its own into the input node when it opens. That charge injection adds to the error and it varies with the input voltage, which makes it a non linearity rather than a fixed offset.

A differential converter injects charge into both inputs, and a matched buffer on each input cancels most of it. The matching of the two buffers and their feedback networks is what determines the residual.

An RC network between the buffer and the converter pin absorbs the charge and limits the current that flows back into the amplifier. The resistor adds its own settling time constant, and the value is a compromise between the two.

source impedance and Loading

source impedance is what the buffer isolates. A sensor with a high output impedance, a divider with large resistors or a filter with a high series resistance all settle slowly, and the buffer removes that dependence.

The buffer’s own input impedance is high and its bias current is small, so it loads the source very little. Its input capacitance is what the source has to drive, and the resulting pole sets the bandwidth of the combination.

A buffer with a rail to rail input and output works from a single supply, and one with a true rail to rail input avoids the crossover distortion that appears in the middle of the range. The requirement follows from the signal range, not from the fashion.

<img src="https://www.gopcba.com/wp-content/uploads/2026/08/pcb1.jpg" alt="ADC input buffer and anti alias filter layout on a measurement PCB” />

anti alias filter and Its Position

An anti alias filter removes the frequencies above half the sampling rate that would otherwise fold back into the measured band. It has to be analog, because the folding happens at the sampling instant.

The filter can be placed before the buffer, between the buffer and the converter, or both. The usual arrangement is a passive RC before the buffer to remove the worst of the broadband noise and a second RC at the converter pin to absorb the charge.

The order of the filter sets its roll off and its phase shift. A single pole is often enough for a slow measurement, and a second or third order is used where the signal has content near the sampling frequency.

Noise and the Buffer’s Own Contribution

The buffer contributes voltage noise, current noise and its own flicker noise. Where the source impedance is high the current noise dominates, and where it is low the voltage noise does.

The noise of the buffer is amplified by nothing, because the buffer has unity gain, and it appears directly at the converter input. Choosing a low noise part for a high resolution converter is therefore worthwhile even though the buffer looks trivial.

Filtering after the buffer reduces both its own noise and the noise that arrives from the source. The corner has to be above the signal band and below the sampling frequency, and it is set from the resolution required.

Layout and Decoupling

Keep the buffer, the RC network and the converter input within a few millimetres. The trace between them carries the charge step, and its inductance and capacitance change the settling behaviour.

Decouple the buffer at its supply pin and keep the return of that capacitor in the analog ground. The charge step flows in the output loop and returns through the ground, so the impedance of that path appears in the measurement.

Keep digital lines and the converter’s own interface away from the analog input. The charge step at the sampling instant couples into any nearby trace, and the coupling is worst where the analog and digital sections run parallel. The layout measures that keep the two apart are described in our guide to mixed signal board design.

Multiplexed Inputs

A multiplexer in front of the converter switches between channels, and the buffer has to settle again after each switch. The stray capacitance of the multiplexer and the leakage of the off channels both add to the problem.

Placing the buffer after the multiplexer means one amplifier has to drive every channel and settle quickly between them. Placing a buffer on each channel costs more amplifiers and removes the settling problem almost entirely.

Verification and Faults

Verify the channel by applying a slow ramp and looking at the difference between the reading and the applied voltage. A bow shaped error across the range indicates a settling or a charge injection problem rather than a converter fault.

Check the reading with the input at the middle of the range and with the converter sampling at its maximum rate. A reading that changes with the rate indicates that the buffer does not settle within the sampling window.

A reading that depends on the previously selected channel is the classic symptom of insufficient settling in a multiplexed system. The release checks that keep such a channel consistent are collected in our PCB design release checklist, the assembly points in judging PCB quality, and the means of controlling noise in our guide to EMI suppression design principles.

FAQ

Do I always need a buffer before an ADC? Not for a low impedance source at a slow sampling rate. The higher the source impedance and the rate, the more necessary it becomes.

Where should the anti alias filter go? Usually an RC before the buffer and a second RC at the converter pin to absorb the sampling charge.

Why does my reading depend on the previous channel? The buffer has not settled between selections. Reduce the source impedance or increase the settling time.