How to choose between discrete and on-board ADCs
by Franck Vermeulen, Senior FAE, Future Electronics (France)
 |
READ THIS TO FIND OUT ABOUT:
 |
- The relative merits of discrete and on-board ADCs
- Which applications benefit from the different types of ADC
|
 |
|
|
Integration has always been one of the main driving factors in
the semiconductor industry, however, one of the hardest
tasks in semiconductor integration is to combine analogue
and digital functions onto a single chip. Even when it can
be achieved, by using high-volume ASICs for example, it
often results in design and performance compromises. Franck
Vermeulen, Senior FAE, Future Electronics (France), explains.
This compromise appears most obviously in the device that stands at
the borderline of the digital and analogue worlds: the Analogue-to-Digital
Converter (ADC). ADCs are often integrated into ICs such as
microcontrollers. However, in contrast to the general rule about
increasing integration, discrete (off-chip) ADCs are forecast to enjoy
strong growth.
So for standard conversion applications, what are the important
differences between an on-chip and off-chip ADC? What should a
designer know when deciding between them?
The issues for the designer fall into three broad categories:
- board-design constraints
- conversion performance
- requirements of the design cycle
Board-design constraints
A microcontroller with an on-chip ADC offers the benefits of reduced
cost and part count compared to a circuit using a discrete ADC.
However, if the signal to be captured (i.e. the sensor) is relatively
remote from the place in which it is to be used (i.e. the data-processing,
control, display or storage components), then the risk of
noise corrupting the signal must be mitigated.
It may, for example, be possible for the converter and the
microcontroller to be relocated next to the sensor. However, this might
be impossible if the application cannot provide enough space around
the sensor to carry the microcontroller and its associated peripheral
components, or if other parts of the complete system prevent the
relocation of the microcontroller. Also, the environment in which the
sensor is located might not be suitable for the microcontroller if, for
instance, the microcontroller could be exposed to extremes of
temperature or hazardous conditions. For these reasons, relocation of
the microcontroller is rarely practical.
The alternative is to adopt circuit-design measures that protect the
integrity of the signal over a long
trace. If the range of the
analogue input signals is large
enough, the noise on the signal
path might have negligible
impact.
Noise can affect not just the
integrity of the signal on the
signal path, but also compromise
the accuracy of the analogue-to-digital
conversion. Microcontrollers
that offer a fully-differential ADC,
such as the MCF52211 ColdFire v2
MCU from Freescale Semiconductor,
help provide immunity from noise.
At their most extreme, ground glitches can reach 200mV in certain
conditions. It is often impossible to split the noisy ground areas from the
analogue signal path when using an integrated ADC, as the signal path and
ground are very close together (see Figures 2 and 3). The result will be output
errors from the ADC. So, in applications with high levels of ground noise,
there is a clear argument in favour of using an external ADC. This allows
complete freedom to optimise the design of the PCB traces for noise
reduction.
Of course, changes to the power-supply design might also reduce the
noise rejected to the ground. The power supply is not the only source of
noise on the ground plane: the power profile of the microcontroller itself will
spill noise to its ground, and hence to the ADC’s input signal. This noise will
bear more relation to CPU usage, than it would to the sampling activity in the
on-board ADC.
The very high frequencies at which today’s microcontrollers operate
generate steep current changes on the power-supply pins of the processor.
Such high di/dt generates voltage spikes in the parasitic inductances on the
circuit. These conditions can, in turn, temporarily change the voltage reference
of the processor, and hence change the reference voltage (Vref) of their
integrated ADCs. This introduces errors into the values captured by them.
The isolated power supply of an off-chip ADC is immune from such noise,
and its performance can often be markedly better than that of an on-chip
ADC.
However, the fundamentals of good board design cannot be ignored:
splitting the power circuits, digital circuits and analogue or small-signal
circuits is often the only way to achieve acceptable input-signal integrity. This
mandates the use of a discrete ADC.

Fig. 1: Example of 3-bit conversion.
Conversion performance and specifications
On-chip ADCs have a performance ceiling that is lower than that of
specialised discrete ADCs. System architects must therefore consider the
required sampling rate and
precision.
To account for the required level
of precision, consider this example:
signal range is 0mV to 40 mV and
required precision is less than
0.4 mV (see Figure 1). A typical
microcontroller is used, offering an
on-chip 12-bit ADC with a Vref of
3V. The design employed uses a
standard pressure sensor such as
the MPX2050 from Freescale. In this
example, the microcontroller’s on-chip
ADC would not be suitable, as
it only provides a 0.73mV precision
value.
To match the requirements of this application with an integrated
ADC, a complete change of microcontroller is needed, to a device
capable of providing an on-chip 14-bit ADC with a Vref of 3V, giving
a precision value of 0.18mV.
Standard microcontrollers with 14-bit ADCs are difficult to find,
especially within a coherent and flexible family. One kind of
microcontroller device that offers high-quality integrated analogue
functionality is the PSoC family from Cypress Semiconductor. PSoC
parts offer the DelSig or ADCINC14 programmable ADCs with 14-bit
precision. These devices can provide a flexible, reconfigurable
combination of analogue functions, but have a small 8-bit CPU that
gives limited controller functionality.
It may, therefore, be possible to provide the necessary functionality with
an integrated converter. To be sure, study some typical microcontroller
datasheets: how many specify the compute bandwidth used by the
converter? The answer is very few. A notable exception is the PSoC family.
See, for instance, AN2239 on www.cypress.com.
Other built-in ADC parameters might also be missing or partly
missing from a standard microcontroller datasheet. For instance:
- What is the Effective Number Of Bits (ENOB)?
- What happens if the signal falls outside the specified input range?
- What is the power consumption at different sampling rates?
- What is the maximum control jitter?
Of course, all of these issues will be clear from the data provided
with any high-precision discrete ADC. Such high-performance ADCs
are available in many variants from suppliers such as National
Semiconductor and austriamicrosystems. These stand-alone ADCs
offers additional advantages:
- Versatile input configurations
- Signal type optimisation
- Support for high-impedance sensors
Finally, the latest generation of stand-alone ADCs offers a way to
gain high precision without the need for a complex signal-conditioning
circuit, or for an expensive sensor with built-in signal
conditioning. These flexible ADCs can zoom and pan to the relevant
portion of the voltage range before they sample it at high resolution.
A good example of this type of flexible ADC is the SX8724 with
ZoomingADC™ technology from Semtech Corporation.

Fig. 2: The effect of noisy ground areas on the analogue signal path.
The requirements of the design cycle
At the outset of the design process it can be hard to know whether
the application might eventually need a faster or more precise ADC
than initially forecast. In this instance, an off-chip device would
appear to offer the flexibility to change the ADC without requiring a
replacement microcontroller.
Indeed, ADC manufacturers often produce families of devices that
make it easy to upgrade or downgrade. National Semiconductor’s
general-purpose ADCs, for instance, offer pin- and code-compatibility
across the whole ADCxx1Sxx1 family.
The PSoC device is another way to build flexibility into the design cycle.
PSoC implements ADC functionality via a set of user-configurable digital
and analogue blocks. These blocks can be easily reconfigured at any
time using Cypress’ PSoC Express or PSoC Designer design tools,
without affecting board layout.
A similar argument applies to products with variants offering
different levels of performance. In this case, again, it is often best to
maintain one basic hardware architecture with one microcontroller,
and change the discrete ADC to meet the requirements of each
product variant.

Fig. 3: The affect of ground glitches that can be as big as 200mV.
Conclusion
This article has described a number of factors that might push a designer
towards choosing a discrete ADC in preference to a microcontroller’s on-board
ADC. These factors include the requirement for flexibility in the
design process, the need for high analogue performance at reasonable
cost, and the constraints of low-noise board design.
The ability of microcontrollers to meet the needs of mixed-signal
designers looks set to become even more constrained in future. The main
market driver for microcontroller manufacturers is the race to offer
improved processor performance and digital feature sets at ever lower
cost. This forces them to migrate constantly to smaller process geometries
to drive down manufacturing cost. State-of-the-art microcontrollers such
as the LPC2468 ARM7 device or the LPC3180 ARM9 device from NXP
Semiconductors are being manufactured on 0.14 micron or 0.09 micron
processes.
By contrast, the best discrete ADCs on the market are manufactured on
processes no smaller than 0.6 microns. So the incompatibility between
the processes for the digital microcontroller circuitry and on-board ADCs
is continuing to grow. This suggests that the discrete ADC has, as the
market data shows, a healthy future ahead of it.