Whether designing a bandgap reference for medical implants, monolithic battery charging or just trying to minimize operating power, a switched capacitor bandgap implementation fits the bill nicely. With a switched cap circuit, the creation of a bandgap voltage is based upon charge transfer as opposed to constantly biased circuitry and thus lends itself well to low power operation. In this implementation, once the bandgap voltage has been created it is stored on capacitors, and thus allows portions of the circuitry to be cycled off for a majority of the time. Storing the bandgap voltage on capacitors in the circuits that use it and then powering down all bandgap reference circuitry can provide low nanowatt average operating power for power sensitive applications.
The following description is divided into three sections, each section represents one of three distinct states of operation. The first state occurs when either the bandgap circuit is first powered up or when it is determined that the output bandgap voltage needs to be refreshed. This refresh is required due to leakage currents slowly drawing charge off the output voltage hold capacitors and hence, degrading bandgap voltage accuracy. In the second state, sampled base-emitter voltages are subtracted to create a PTAT (Proportional To Absolute Temperature) charge component which is added to a sampled, single base-emitter voltage which has a CTAT (Complementary To Absolute Temperature) charge component. When these two components are summed in correct proportion to each other they produce a 0TC (0 Temperature Coefficient) output voltage on operational transconductance amplifier (OTA) feedback capacitors. In the third state, the 0 TC output voltage (V0TC) is held on these feedback capacitors, typically for a long time relative to the time of states 1 and 2 thus minimizing operating power.
STATE 1
Figure1 shows the circuit in its initial state with all φ1 and φ3 switches closed and all φ2 switches open. In this state, neglecting the OTA offset voltage, feedback capacitors C´´ are both shorted and the differential output voltage, VOUT , is 0 volts. Capacitors labeled C sample VBE1 and ground and capacitors labeled C´ sample both VBE0 and VBE1 . Not shown but assumed in all figures is a common mode feedback circuit that keeps both OTA outputs centered at mid-supply (vcm). This mid-supply voltage is also used as a common connection point seen in all figures.
STATE 2
Figure2 shows the second state of operation where the sampled voltages from state 1, which are saved as charge on the sampling capacitors, is transferred onto the OTA feedback capacitors, C´´.
In this state all φ2 and φ3 switches are closed and all φ1 switches open. The charges stored in state 1 and transferred in state 2 are as follows:

Hence, the OTA output voltage is:

Note that the doubling of the PTAT charge component in the above equation is accomplished by switching the C´ sampling caps between VBE0 and VBE1 instead of to vcm in this state. This is done to halve the size of both C´ capacitors. The ratio of C to C´ is set such that VOUT has a zero temperature coefficient (0TC) and C´ ´is chosen to scale VOUT to the desired output level (V0TC ).
STATE 3
Figure 3 shows the third and last state of operation in which V0TC , developed at the OTA output in state 2, is held on the OTA feedback capacitors by opening the φ3 switches. All φ1 and φ2 switches retain their state 2 positions. In this state, the OTA input is disconnected from the input bipolar transistor circuitry so it can be powered off to reduce operating power. In addition, when the φ3 switches are opened, the closed loop bandwidth of the OTA is increased since this puts the OTA into a unity gain configuration. This allows faster response time at lower OTA power when the output is used to drive other switched cap circuitry such as a sigma delta ADC.
In conclusion, the use of switched capacitor techniques for generating bandgap based voltages has many benefits. Among these are low operating power, flexible output voltage scaling and compatibility with other switched capacitor circuits. In this article, the basic functioning of this type of circuit has been explained without reference to accuracy. How to leverage this architecture to minimize the errors associated with typical bandgap circuits will be explained in part 2 of this series.
Scott:
The switched-cap bandgap reference goes back at least
twenty years, and there are several patents which show
various possibilities, including one from Analog Devices:
Patent 5,563,504 Switching Bandgap Voltage Reference
Oct. 8, 1996 (Gilbert & Shu)
Ours differs from what is described in this Planet Analog
article in being much simpler in implementation: it uses
only ONE bipolar transistor – a substrate PNP of course –
and fewer switches.
In Phase 1, it receives a first current (which need not be
especially accurate, since every band-gap reference must
be trimmed to value) and this first value of VBE is stored
in a capacitor.
In Phase 2 the first drive current is now augmented by a
second additional current, say 100 times larger. (Ratios,
of course, can optionally be made to high accuracy). In
this example, the current ratio will be 101, and the VBE
will, ideally, increase by 119.3mV at 300K, providing the
basic PTAT component.
Again, the precise value of this delta-VBE is not critical,
since the final reference voltage must be trimmed to an
appropriate target value. However, to avoid running into
errors caused by contact resistances in the PNP – which
would have an aberrant temperature- profile — both of
these currents are kept in the micro- to nanoamp range.
It remains only to add this delta-VBE to the earlier VBE
in the simple switched capacitor amplifier, now using a
capacitor ratio that raises it by a factor of roughly five
to make ~600mV PTAT + (say) 600mV CTAT, getting
us to the magic number of (around) 1.2V.
Of course, if the reference is needed continuously, one
needs to add a final sample/hold stage. But there are a
surprising number of actual systems which only require
the reference to be available on-demand and briefly.
At very low bias levels, nonlinear charge transfer in the
switches may become problematical.
Barrie Gilbert