22 Digital Calibration of Analog Circuits and Systems
is not critical, autozero is also suitable (see section 4.3 for details). In terms of
1/f noise removal, they are both comparable.
On the other hand, when a high signal bandwidth is necessary, autozero is
much more adapted because it keeps the signal in the baseband instead of
modulating it at higher frequencies.
The two last columns of table 1 present the characteristics regarding the
compensation frequency. If the system is calibrated frequently, all solutions
are adapted. If on the other hand the calibration is performed at long time
intervals, the autozero using digital compensation circuitry is much easier to
implement. This is due to the fact that analog memories have a limited reten-
tion time and need to be refreshed periodically.
There exist analog storage techniques like floating gate which allow
extremely long retention times [18], but they imply extra processing steps,
specific devices and/or high programming voltages [19]. On the other hand,
storing digital information over a long period of time is straightforward. The
only limitation is the data loss due to power off.
Another approach is factory trimming, which is one time programming. It
can compensate for initial imperfections, but does not correct drifts due to cir-
cuit ageing or temperature variations. This approach is somewhat different
from chopper and autozero, but it can in some cases be sufficient.
The compensation frequency depends on the factors that affect the varia-
tions of system parameters. If the circuit is very stable, a one time factory
trimming can suffice. If the characteristics change with the age of the circuit,
a single calibration at power-up can be considered. If they change with tem-
perature, calibration can be performed periodically at low frequency (in the
Hz range). Finally, if the variation is faster, like 1/f noise for instance, the
compensation must also be performed at higher rates.
9 CONCLUSION
There are various techniques to improve the performances of analog cir-
cuits. Although they all have specific applications, the digital solutions
generally seem better adapted to the long-term evolution of manufacturing
processes. With the ever more shrinking of devices, it becomes more and
more difficult to design intrinsically high-performance analog circuits. If dig-
ital correction circuits are available to compensate for the imperfections of
analog components, it will still be possible to design high-performance cir-
cuits with future deep sub-micron fabrication processes. Furthermore, if the
digital correction circuits and algorithms are systematic, it will be possible to
automate their design and allow the designers to focus their attention on the
analog blocks.