I'm a little envious: I see lots of attention given to processors, system architectures, IoT wireless connectivity, development software and tools, GUIs, embedded processors, ultra-low power anything and everything – you catch my drift. Yet for many of their intended applications, success starts with a sensor and its signal conditioning front end (often called the AFE—analog front end). In some cases, such as a basic temperature measurement to ±0.5⁰C, you can get away cheap and easy by using a diode or thermistor and a slow but high-resolution A/D converter that is in the microcontroller itself.
But that’s just for some cases. Today's higher-end instrumentation needs a good sensor, and equally important, a correspondingly good AFE. You can dress up the GUI with as many bells and whistles as you want, but if the sensor is inadequate or the AFE doesn’t do it justice, you've got a pretty picture and not much else. Even more numeric processing won't help because, in most cases, the algorithms can’t improve the basic accuracy and precision of the transducer and its AFE (although they may be able to do something about noise by averaging or curve fitting).
The AFE problem has become more challenging in recent years, for several reasons. First, users are demanding increased accuracy in measurements, with parts per billion (ppb) resolution replacing parts per million (ppm). Second, many of these instruments are in unattended, remote, or outdoor locations, so temperature drift of the electronics is a major factor. Third, some of the most exciting and innovative sensing design involves subtle optical phenomena and transducers such as for Raman spectroscopy, and the AFE must be matched to the unique and often complex characteristics of these devices.
Ironically, cost is not the major factor in many of these designs (how often do you hear that?) despite the standard disclaimer "we're in a very cost-competitive market." For many of these instruments, the total BOM value is quite high, and another few dollars on the AFE to ensure leading-edge performance is a good use of the budget. Further, many of them have just one or two sensor channels, so the actual impact a higher-price AFE is moderated (yes, there are exceptions to that statement, such as esoteric physics research at CERN's Large Hadron Collider or the IceCube Telescope at the South Pole).
Next: Don't ignore temp.