Charge pump with temporally-varying adiabaticity

This application is a continuation of U.S. application Ser. No. 14/719,815, filed on May 22, 2015, which claims the benefit of the priority date of U.S. application Ser. No. 14/027,716, filed on Sep. 16, 2013, issued as U.S. Pat. No. 9,041,459 on May 26, 2015, the contents of which are hereby incorporated by reference in their entiretyNotice: More than one reissue application has been filed for the reissue of U.S. Pat. No. 10,162,376. The reissue applications are Reissue application Ser. No. 17/133,909, and the present application filed herewith. This application is a divisional reissue application of U.S. patent application Ser. No. 17/133,909, filed Dec. 24, 2020, which is a broadening reissue of U.S. Pat. No. 10,162,376, issued on Dec. 25, 2018, from U.S. patent application Ser. No. 15/460,596, which is a continuation of U.S. application Ser. No. 14/719,815, filed on May 22, 2015, which claims the benefit of the priority date of U.S. application Ser. No. 14/027,716, filed on Sep. 16, 2013, issued as U.S. Pat. No. 9,041,459 on May 26, 2015, the contents of which are hereby incorporated by reference in their entirety.

This invention relates to adiabatic power conversion, and in particular to configuration and control for partial adiabatic operation of a charge pump.

Various configurations of charge pumps, including Series-Parallel and Dickson configurations, rely on alternating configurations of switch elements to propagate charge and transfer energy between the terminals of the charge pump. Energy losses associated with charge propagation determine the efficiency of the converter.

1 FIG. 1 FIG. 100 110 190 110 190 100 Referring to, a single-phase Dickson charge pumpis illustrated in a step-down mode coupled to a low-voltage loadand a high-voltage source. In the illustrated configuration, generally the low-voltage loadis driven (on average) by a voltage that is ⅕ times the voltage provided by the source and a current that is five times the current provided by the high-voltage source. The pump is driven in alternating states, referred to as state one and state two, such that the switches illustrated inare closed in the indicated states. In general, the duration of each state is half of a cycle time T and the corresponding switching frequency of the charge pumpis equal to the inverse of the cycle time T.

2 FIGS.A-B 1 4 100 190 100 1 4 illustrate the equivalent circuit in each of states two and state one, respectively, illustrating each closed switch as an equivalent resistance R. Capacitors Cthrough Chave a capacitance C. In a first conventional operation of the charge pump, the high-voltage sourceis a voltage source, for example, a twenty-five volt source, such that the low-voltage loadis driven by five volts. In operation, the voltage across the capacitors Cthrough Care approximately five volts, ten volts, fifteen volts, and twenty volts, respectively.

100 2 1 4 1 110 3 2 4 1 110 2 FIGS.A-B 2 FIG.A One cause of energy loss in the charge pumprelates the resistive losses through the switches (i.e., through the resistors R in). Referring to, during state two, charge transfers from the capacitor Cto the capacitor Cand from the capacitor Cto the capacitor C. The voltages on these pairs of capacitors equilibrate assuming that the cycle time T is sufficiently greater than the time constant of the circuit (e.g., that the resistances R are sufficiently small. Generally, the resistive energy losses in this equilibration are proportional to the time average of the square of the current passing between the capacitors and therefore passing to the low-voltage load. Similarly, during state one, the capacitors Cand Cequilibrate, the capacitor Ccharges, and the capacitor Cdischarges, also generally resulting in a resistive energy loss that is proportional to the time average of the square of the current passing to the low-voltage load.

110 For a particular average current passing to the load, assuming that the load presents an approximately constant voltage, it can be shown than the resistive energy loss decreases as the cycle time T is reduced (i.e., switching frequency is increased). This can generally be understood by considering the impact of dividing the cycle time by one-half, which generally reduces the peak currents in the equilibration by one half, and thereby approximately reduces the resistive energy loss to one quarter. So the resistive energy loss is approximately inversely proportional to the square of the switching frequency.

However, another source of energy loss relates to capacitive losses in the switches, such that energy loss grows with the switching frequency. Generally, a fixed amount of charge is lost with each cycle transition, which can be considered to form a current that is proportional to the switching frequency. So this capacitive energy loss is approximately proportional to the square of the switching frequency.

Therefore, with a voltage source and load there an optimal switching frequency that minimizes the sum of the resistive and capacitive energy losses, respectively reduced with increased frequency and increased with increased frequency.

1 2 FIGS.andA 2 FIG.A 110 Patent Publication WO 2012/151466, published on Nov. 8, 2012, describes configurations in which the source and/or load comprise regulating circuits. In particular, in-B, the loadcan effectively comprise a current sink rather than present a constant voltage in an example of what is referred to as “adiabatic” operation of a charge pump. If the current sink accepts constant current, then the currents illustrated ineffectively remain substantially constant values during the illustrated state. Therefore, the resistive power loss is lower than the resistive loss in the voltage driven case discussed in the Background, and also substantially independent of the cycle time T. In situations in which the load sinks a pulsed current, then for a particular average current, the resistive energy loss generally increases as the duty cycle of the current decreases (and the peak current increases). There is a range of low duty cycles in which the resistive losses with a pulsed current exceed the losses for the same average current that would result from the charge pump driving a relatively constant output voltage, for example, across a large output capacitor.

In one aspect, in general, operation of a charge pump is controlled to optimize power conversion efficiency by using an adiabatic mode with some operating characteristics and a non-adiabatic mode with other characteristics. The control is implemented by controlling a configurable circuit at the output of the charge pump.

In another aspect, in general, operation of a charge pump is controlled so that resistive power losses are minimized by using an adiabatic mode with relatively high duty cycle (i.e., relatively high output current) and using a non-adiabatic mode with relative low duty cycle (e.g., relatively low output current). In some examples, mode is selected by selectively introducing a compensation capacitor at the output of the charge pump to present a substantially constant voltage.

In another aspect, in general, an apparatus a charge pump and a controller coupled to the charge pump. The charge pump has a plurality of switch elements arranged to operate in a plurality cycles, with each cycle being associated with a different configuration of the switch elements. The switch elements are configured to provide charging and discharging paths for a plurality of capacitive elements. The controller has an output for controlling timing of the cycles of the charge pump and one or more sensor inputs for accepting sensor signals charactering operation of the charge pump and/or operation of peripheral circuits coupled to the charge pump. The controller is configured adjust the timing of the cycles of the charge pump according variation of the one or more sensor inputs within cycles of operation of the charge pump.

In another aspect, in general, an apparatus includes a switched capacitor charge pump configured to provide a voltage conversion between terminals including a high voltage terminal and a low voltage terminal. The apparatus also includes a compensation circuit coupled to a first terminal of the charge pump for driving a load by the charge pump, the compensation circuit providing a capacitance configurably couplable to the first terminal of the charge pump. A controller is coupled to charge pump and the configurable circuit, and has an output for configuring the compensation circuit, and one or more sensor inputs for accepting sensor signals charactering operation of the charge pump and/or operation of peripheral circuits coupled to the charge pump. The controller is configured to configure the compensation circuit according to the sensor signals to affect efficiency of power conversion between a power source coupled to the charge pump and the load coupled to the charge pump via the configurable circuit.

Aspects may include one or more of the following features.

The controller is configured to couple a selected capacitance to the first terminal to optimize an efficiency of the power conversion.

The one or more sensor signals include a sensor signal that characterizes time variation of a current passing to or from the charge pump via the compensation circuit. In some examples, the sensor signal characterizes a duty cycle of a pulsed current passing to or from the charge pump. In some examples, this current passing to or from the charge pump via the compensation circuit is a current passing between the compensation circuit and a peripheral coupled to the charge pump via the compensation circuit.

The one or more sensor signals include a sensor signal that characterizes a voltage at at least one of the terminals of the charge pump and at the peripheral circuit coupled to the charge pump.

The one or more sensor signals include a sensor signal that characterizes switching frequency of the charge pump.

The controller is configured to determine an operating mode from the sensor signals, and to determine the configuration of the compensation circuit according to the determined mode.

The controller is configured to identify at least a mode having fast switching limit operation of the charge pump and a pulsed current load, and increase the capacitance coupled to the first terminal in said mode.

The controller is configured to identify at least a mode having slow switching limit operation of the charge pump and a pulsed current load with a duty cycle less than a threshold duty cycle, and increase the capacitance coupled to the first terminal in said mode.

The apparatus further includes a peripheral circuit that includes a regulator coupled to the compensation circuit. The regulator provides a current-based load via the compensation circuit to charge pump. The controller is configured to determining a configuration of the compensation circuit according to an efficiency of power conversion performed by the charge pump. In some examples, the regulator comprises a Buck converter. In some examples, the charge pump comprises a Series-Parallel charge pump. In some examples, the charge pump comprises a Dickson charge pump.

In another aspect, in general, a method is directed to power regulation using a charge pump coupled to a load using a compensation circuit coupled to a terminal of the charge pump. The method includes configuring a capacitance provided by the compensation circuit to a first terminal of the charge pump. The capacitance is selected according to the sensor signals to affect efficiency of power conversion between a power source coupled to the charge pump and the load coupled to the charge pump via the configurable circuit.

The method may include acquiring the sensor signals. The sensor signals may characterize one or more a time variation of a current passing to or from the charge pump via the compensation circuit, a duty cycle of a current passing between the compensation circuit and a peripheral circuit, a voltage at the first terminal of the charge pump, and a voltage at the peripheral circuit coupled to the charge pump.

One advantage of one or more embodiments is that efficient operation is maintained in varying operating modes of the power converter.

Another advantage of one or more embodiments is that a controller does not have to be preconfigured for a particular use of a charge pump and can adapt to the circuit in which the pump is embedded without further configuration. For example, the controller can adapt to the size of pump capacitors used, type of regulator coupled to the pump, switching frequency of the pump and/or regulator, etc.

Other features and advantages of the invention are apparent from the following description, and from the claims.

100 110 190 110 1 FIG. 1 2 FIGS.andA 2 FIG.A As introduced above, as one example, a charge pumpillustrated inmay be operated in an “adiabatic” mode in which one or both of a low-voltage peripheraland a high-voltage peripheralmay comprise a current source. For example, Patent Publication WO 2012/151466, published on Nov. 8, 2012, and incorporated herein by reference, describes configurations in which the source and/or load comprise regulating circuits. In particular, in-B, the low-voltage loadcan effectively comprise a current source rather than a voltage source in an example of what is referred to as “adiabatic” operation of a charge pump. If the current source maintains a constant current from the charge pump, then currents illustrated inmaintain substantially constant values during the illustrated state. Therefore, the resistive losses in the switches through which the current passes are lower than the resistive loss in the voltage load case, and also substantially independent of the switching frequency and the cycle time T. As in the voltage driven case, there capacitive losses in the switches grow with increasing switching frequency, which suggests that lowering the switching frequency is desirable. However, other factors, which may depend on internal aspects of the charge pump, voltage or current characteristics at the terminals of the charge pump, and/or internal aspects of the peripheral elements, such as the source and/or load, may limit the cycle time (e.g., impose a lower limit on the switching frequency).

3 FIG. 2 FIGS.A-B 320 312 320 320 100 100 Referring to, in a first mode of operation, a loadcan be considered to comprise a constant current sourcewith an output current IO. In some implementations, the loadalso includes an output capacitor, which for the analysis below can be considered to be small enough such that current passing to the loadcan be considered to be substantially constant. As introduced above with reference to, the charge transfer between capacitors in the charge pumpduring the alternating states of operation of the charge pumpare therefore substantially constant in the adiabatic mode of operation.

3 FIG. 340 100 320 344 342 100 Continuing to refer to, a compensation circuitis introduced between the charge pumpand the load. A switchis controllable to selectively introduce a compensation capacitorto the output of the charge pump.

3 FIG. 392 100 10 100 342 344 350 344 350 Various factors can affect the efficiency of the power conversion illustrated in, including the voltage of an input voltage source, the switching frequency of the charge pump, and the output current(or somewhat equivalently the input or output current of the charge pump). The efficiency is also dependent on whether or not the compensation capacitoris coupled to the output path via the switch. As a general approach, a controlleraccepts inputs that characterize one or more factors that affect efficiency and outputs a control signal that sets the state of the switchaccording to whether efficiency is expected to be improved introducing the compensation capacitor versus not. A further discussion of logic implemented by the controlleris provided later in this Description.

4 FIG. 4 FIG. 100 320 340 100 392 100 320 322 324 326 328 322 324 322 324 100 322 320 100 320 100 320 100 320 100 320 Referring to, in another example, a configuration of a charge pumphas a regulatorcoupled via a compensation circuitto the low-voltage terminal of a charge pump, and a voltage sourcecoupled to the high-voltage terminal of the charge pump. The regulator(also referred to below generally interchangeably as a “converter”) illustrated inis a Buck converter, which consists of switches,, an inductor, and an output capacitor. The switches open and close (i.e., present high and low impedance, respectively) in alternating states, such that the switchis open when then the switchis closed, and the switchis closed when the switchis open. These switches operate at a frequency than can be lower, higher, or equal to the switches in the charge pump, with a duty cycle defined as the fraction of time that the switchin the regulatoris closed. A preferred embodiment is when the switching frequency of the charge pumpis lower than the regulator. However, in the case the charge pumpis at a higher frequency than the regulator, the charge-pumpis disabled when the regulatoris off (low duty cycle) and the charge-pumpis enabled when the regulatoris on.

320 320 322 100 320 326 320 100 100 In general, the regulatoroperates at its highest power efficiency when it operates at its highest duty cycle. In some examples, a controller of the regulator (not shown) adjusts the duty cycle in a conventional manner to achieve a desired output voltage VO. During the cycles of the regulatorin which the switchis closed, the current passing from the charge pumpto the regulatoris effectively constant, equal to the current through the inductor. Assuming that the switching frequency of the regulatoris substantially higher than the switching frequency of the charge pump, the charge pumpcan be considered to be driven by a pulsed current source with an average current equal to the duty cycle times the inductor current.

320 100 350 344 342 100 100 350 100 Note that as introduced above, in situations in which the regulatorsinks a pulsed current, then for a particular average current, the resistive energy loss generally increases as the duty cycle of the current decreases, approximately inversely with the duty cycle. There is a range of low duty cycles, and thereby high peak current relative to the average current, in which the resistive losses with a pulsed current exceed the losses for the same average current that would result from the charge pumpdriving a relatively constant output voltage, for example, across a large output capacitor. Therefore, for a selected range of low duty cycles, the controllercloses the switchand introduces a relatively large compensation capacitorat the output of the charge pump. The result is that the charge pumpis presented with a substantially constant voltage, and therefore operates in a substantially “non-adiabatic” mode. Therefore, the controlleris effectively responsive to the output voltage because the duty cycle is approximately proportional to the output voltage. Thereby operating the charge pumpin an adiabatic mode at high output voltage and in a non-adiabatic mode at low output voltage; and switches between the adiabatic and non-adiabatic modes at a threshold duty cycle to maintain an optimum efficiency of the overall power conversion.

350 4 5 FIGS.and Examples of control logic implemented in the controllerin configurations such as those illustrated incan be under in view of the following discussion.

In general, a charge pump can operate in one of two unique operating conditions, or in the region in between them. In a slow switching limit (SSL) regime the capacitor currents in the charge pump have the time to settle to their final values and capacitor voltages experience significant change in magnitude from beginning to end of a cycle of the charge pump operation. In the fast switching limit (FSL) regime, the capacitors do not reach equilibrium during a cycle of the charge pump operation, for instance, due to a combination of one or more of high capacitances, high switching frequency, and high switch resistances.

100 344 342 100 320 4 FIG. Another factor relates to the capacitance at the output of the charge pump, which in the circuits ofcan be increased by closing the switchto add the compensation capacitorto the output. For small output capacitance, the output current of the charge pumpis effectively set by the pulsed current characteristic of the regulator. As discussed above, for a given average current, the resistive power losses in the pulsed current case are approximately inversely proportional the duty cycle.

100 100 342 320 100 For large output capacitance, the RMS of the output current of the charge pumpis effectively determined by the equilibration of the internal capacitors of the charge pumpwith the compensation capacitorand the regulator. For a given average current, this resistive power loss is approximately inversely proportional to the square of the peak-to-peak voltage across the internal capacitors in the charge pump.

342 3 4 FIGS.and Four combinations of FSL/SSL and constant/pulsed IO modes of operation are possible. In some examples, each of these four modes is affected in different ways based on the addition of a compensation capacitoras shown in.

3 FIG. 342 Case one: In FSL mode, with constant output current IO as in, introduction of the compensation capacitordoes not substantially affect conversion efficiency.

4 FIG. 342 100 Case two: In FSL mode with pulsed output current as in, efficiency increases when the compensation capacitoris introduced, thereby reducing the RMS current seen by the charge pump.

3 FIG. 342 Case three: In SSL mode, with constant output current IO as in, efficiency generally increases without introduction of the compensation capacitor, thereby yielding adiabatic operation.

4 FIG. 100 342 342 100 Case four: In SSL mode, with pulsed load current as in, efficiency depends on the relation between the average output current, the duty cycle, and how far the charge pumpis operating from the SSL/FSL boundary. For example, at low duty cycle, efficiency generally increases with introduction of the compensation capacitor, thereby yielding non-adiabatic operation. In contrast, at high duty cycle, efficiency generally increases without introduction of the compensation capacitor, thereby yielding adiabatic operation. Furthermore, when the charge pumpis in SSL mode, the farther from the SSL/FSL boundary, the lower the duty cycle at which the efficiency trend reverses.

Depending on the relative values of charge pump capacitors, switch resistances and frequency, it is possible that the charge pump operate in a regime between FSL and SSL. In this case, there is effectively a transition point between case four and case two at which the compensation capacitor is introduced according to the overall efficiency of the conversion. As described above, knowledge of the average charging current and its duty cycle is necessary in case four for determining if introduction of the compensation capacitor will improve efficiency.

350 In some implementations, the controllerdoes not have access to signals or data that directly provide the mode in which the power conversion is operating. One approach is for the controller to receive a sensor signal that represents the input current of the charge pump, and infer the operating mode from that sensor signal.

109 4 1 FIG. As an example, a sensor signal determined as a voltage across the switch at the high voltage terminal of the converter (e.g., the switch between sourceand the capacitor Cin) can be used to represent the current because when the switch is closed, the voltage is the current times the switch resistance.

5 FIG. 510 530 540 An alternative circuit shown inprovides a scaled version of the input current IIN. The input switch, with closed resistance R is put in parallel with a second switch with closed resistance kR, for example, fabricated as a CMOS switch where the factor k depends on the geometry of the switch. When the switches are closed the differential amplifiercontrols the gate voltage of a transistorsuch that the voltage drop across the two switches are equal, thereby yielding the scaled input current IIN/k, which can be used to form a sensor input signal for the controller.

The sensed input current can be used to determine whether the compensation capacitor should be switched in, for example, according to a transition between case four and case two described above.

100 One possible method for determining the operation mode of the charge pumpconsists of taking two or more measurements of the input current IIN and establishing that the difference between the values of consecutive samples is substantially zero for SSL mode, or is above a pre-determined threshold for FSL mode.

100 350 350 100 350 100 350 344 342 Another method is to measure the difference in the voltage of a capacitor in the charge pump. Once the input current IIN is known, the controllercan infer the operating mode based upon the voltage ripple on the capacitor over a full cycle. Note that the controllerdoes not necessarily know the particular sizes of capacitors that are used in the charge pump, for example, because the capacitors are discrete capacitors that are not predetermined. However, the capacitor values can be inferred from knowledge of the current, voltage ripple, and frequency, thereby allowing the controllerto determine whether the charge pumpis operating in the FSL or SSL mode. The controllercan then select adiabatic or non-adiabatic charging by controlling the switchto selectively introduce the compensation capacitor.

Other controller logic is used in other implementations. For example, an alternative is for the controller to measure efficiency given by:η=VO/(N*VIN)where η is the efficiency, VO is the measured converter output voltage, VIN is the measured converter input voltage, and N is the charge pump conversion ratio.

The controller directly measures the effect of selecting adiabatic vs. non-adiabatic charging on converter efficiency by comparing the average value of the output voltage VO over a complete charge pump cycle.

Other controller logic uses combinations of the approaches described above. For instance, the controller can confirm that the assessment of charge pump operating mode and estimation of efficiency increase by changing the charge pump charging mode.

100 100 1 1 100 1 100 100 100 1 FIG. 6 FIG. A traditional method for operating the charge pumpis at a fixed frequency in which the switching occurs independently of the load requirement (i.e., the switches inoperate on a fixed time period). Referring to, during one cycle of the switching of the charge pump, a current Idischarges from the capacitor Cand a current IP discharges other of the capacitors in the charge pump. For a particular intermediate current IX, the longer the cycle time T, the larger the drop in voltage provided by the capacitor C. A consequence of this is that the switching frequency generally limits the maximum intermediate current IX because the switching frequency for a particular load determines the extent of voltage excursions, and in some cases current excursions (i.e., deviations, variation), at various points and between various points within the charge pumpand at its terminals. For a particular design of charge pump, or characteristics of load and/or source of the charge pump, there are operational limits on the excursions.

7 FIGS.A-C 7 FIG.A 7 FIG.A 7 FIGS.A-B 100 320 Referring to, the intermediate voltage VX of the charge pumpis shown in various current and timing examples. Referring to, at a particular intermediate current IX, the intermediate voltage VX generally follows a saw-tooth pattern such that it increases rapidly at the start of each state, and then generally falls at a constant rate. Consequently, the rate of voltage drop depends on the output current IO. At a particular output current IO and switching time, a total ripple voltage δ results, and a margin over the output voltage VO is maintained, as illustrated in. (Note that the graphs shown indo not necessarily show certain features, including certain transients at the state transition times, and related to the high frequency switching of the regulator; however these approximations are sufficient for the discussion below).

7 FIG.B 4 FIG. 316 320 320 Referring to, in the output current IO in the circuit inincreases, for instance by approximately a factor of two, the ripple of the intermediate voltage VX increases, and the minimum intermediate voltage VMIN decreases and therefore for a constant output voltage VO the margin (i.e. across inductor) in the regulatordecreases. However, if the voltage margin decreases below a threshold (greater than zero), the operation of the regulatoris impeded.

7 FIG.C 7 FIG.A 320 Referring to, to provide the regulatorwith a sufficient voltage margin voltage the switching frequency can be increases (and cycle time decreased), for example, to restore the margin shown in. Generally, in this example, doubling the switching frequency compensates for the doubling of the output current IO. However more generally, such direct relationships between output current IO or other sensed signals and switching frequency are not necessary.

100 100 100 In general, a number of embodiments adapt the switching frequency of the charge pumpor determine the specific switching time instants based on measurements within the charge pumpand optionally in the low-voltage and/or high-voltage peripherals coupled to the terminals of the charge pump.

4 FIG. 350 100 100 In a feedback arrangement shown in, the controlleradapts (e.g., in a closed loop or open loop arrangement) the switching frequency. For any current up to a maximum rated current with a fixed switching frequency, the charge pumpgenerally operates at a switching frequency lower than (i.e., switching times greater than) a particular minimum frequency determined by that maximum rated current. Therefore, when the current is below the maximum, capacitive losses may be reduced as compared to operating the charge pumpat the minimum switching frequency determined by the maximum rated current.

100 100 320 326 320 One approach to implementing this feedback operation is to monitor the intermediate voltage VX and adapt operation of the charge pump to maintain VMIN above a fixed minimum threshold. One way to adapt the operation of the charge pumpis to adapt a frequency for the switching of the charge pumpin a feedback configuration such that as the minimum intermediate voltage VMIN approaches the threshold, the switching frequency is increased, and as it rises above the threshold the switching frequency is reduced. One way to set the fixed minimum threshold voltage is as the maximum (e.g., rated) output voltage VO of the regulator, plus a minimum desired margin above that voltage. As introduced above, the minimum margin (greater than zero) is required to allow a sufficient voltage differential (VX−VO) to charge (i.e., increase its current and thereby store energy in) the inductorat a reasonable rate. The minimum margin is also related to a guarantee on a maximum duty cycle of the regulator.

320 320 100 100 A second approach adapts to the desired output voltage VO of the regulator. For example, the regulatormay have a maximum output voltage VO rating equal to 3.3 volts. With a desired minimum margin of 0.7 volts, the switching of the charge pumpwould be controlled to keep the intermediate voltage VX above 4.0 volts. However, if the converter is actually being operated with an output voltage VO of 1.2volts, then the switching frequency of the charge pumpcan be reduced to the point that the intermediate voltage VX falls as low as 1.9 volts and still maintain the desired margin of 0.7 volts.

312 314 In a variant of the second approach, rather than monitoring the actual output voltage VO, an average of the voltage between the switches,may be used as an estimate of the output voltage VO.

100 In yet another variant, the switching frequency of the charge pumpis adapted to maintain the intermediate voltage VX below a threshold value. For example, the threshold can be set such that the intermediate voltage VX lowers or rises a specific percentage below or above the average of the intermediate voltage VX (e.g. 10%). This threshold would track the intermediate voltage VX. Similarly, a ripple relative to an absolute ripple voltage (e.g. 100 mV) can be used to determine the switching frequency.

100 Note also that the voltage ripple on the output voltage VO depends (not necessarily linearly) on the voltage ripple on the intermediate voltage VX, and in some examples the switching frequency of the charge pumpis increased to reduced the ripple on the output voltage VO to a desired value.

100 1 4 100 Other examples measure variation in internal voltages in the charge pump, for example, measuring the ripple (e.g., absolute or relative to the maximum or average) across any of the capacitors Cthrough C. Such ripple values can be used instead of using the ripple on the intermediate voltage VX in controlling the switching frequency of the charge pump. Other internal voltages and/or currents can be used, for example, voltages across switches or other circuit elements (e.g., transistor switches), and the switching frequency can be adjusted to avoid exceeding rated voltages across the circuit elements.

320 350 100 320 326 100 320 100 320 In addition to the desired and/or actual output voltages or currents of the regulatorbeing provided as a control input to the controller, which adapts the switching frequency of the charge pump, other control inputs can also be used. One such alternative is to measure the duty cycle of the regulator. Note that variation in the intermediate voltage VX affects variation in current in the Buck converter's inductor. For example, the average of the intermediate voltage VX is generally reduced downward with reducing of the switching frequency of the charge pump. With the reduction of the average output voltage VO, the duty cycle of the regulatorgenerally increases to maintain the desired output voltage VO. Increasing the duty cycle generally increases the efficiency of a Buck converter. So reducing the switching frequency of the charge pumpcan increase the efficiency of the regulator.

It should be understood that although the various signals used to control the switching frequency may be described above separately, the switch frequency can be controlled according to a combination of multiple of the signals (e.g., a linear combination, nonlinear combination using maximum and minimum functions, etc.). In some examples, an approximation of an efficiency of the charge pump is optimized.

350 100 350 100 100 100 6 FIG. The discussion above focuses on using the controllerto adjust the switching frequency of the charge pumpin relatively slow scale feedback arrangement. The various signals described above as inputs to the controllercan be used on an asynchronous operating mode in which the times at which the charge pumpswitches between cycles is determined according to the measurements. As one example, during state one as illustrated in, the intermediate voltage VX falls, and when VX−VO reaches a threshold value (e.g., 0.7 volts), the switches in the charge pumpare switched together from state one to state two. Upon the transition to state two, the intermediate voltage VX rises and then again begins to fall, and when VX−VO again reaches the threshold value, the switches in the charge pumpare switched together from state two back to state one.

In some examples, a combination of asynchronous switching as well as limits or control on average switching frequency for the charge pump are used.

100 344 342 342 Unfortunately, as the intermediate current IX decreases the switching frequency of the charge pumpdecreases as well. This can be problematic at low currents because the frequency could drop below 20 kHz, which is the audible limit for human hearing. Therefore, once the frequency has dropped below a certain limit, a switchcloses and introduces a compensation capacitor. This forces the converter into non-adiabatic operation allowing the frequency to be fixed to a lower bound (e.g. 20 kHz). Consequently, the compensation capacitoris introduced when either the duty cycle is low or when the output current IO is low.

Note that the examples above concentrate on a compensation circuit that permits selectively switching a compensation capacitor of a certain fixed capacitance onto the output of the charge pump. More generally, a wide variety of compensation circuits can be controlled. One example is a variable capacitor, which can be implemented as a switched capacitor bank, for example, with power of two capacitances. The optimal choice of capacitance generally depends on the combination of operating conditions (e.g., average current, pulsed current duty cycle, etc.) and/or circuit configurations (e.g., type of regulators, sources, load, pump capacitors), with the determining of the desired capacitance being based on prior simulation or measurement or based on a mechanism that adjusts the capacitance, for instance, in a feedback arrangement. In addition, other forms of compensation circuits, for example, introducing inductance on the output path, networks of elements (e.g., capacitors, inductors).

Note that the description focuses on a specific example of a charge pump. Many other configurations of charge pumps, including Dickson pumps with additional stages or parallel phases, and other configurations of charge pumps (e.g., series-parallel), can be controlled according to the same approach. In addition, the peripherals at the high and/or low voltage terminals are not necessarily regulators, or necessarily maintain substantially constant current. Furthermore, the approaches described are applicable to configurations in which a high voltage supply provides energy to a low voltage load, or in which a low voltage supply provides energy to a high voltage load, or bidirectional configurations in which energy may flow in either direction between the high and the low voltage terminal of the charge pump. It should also be understood that the switching elements can be implemented in a variety of ways, including using Field Effect Transistors (FETs) or diodes, and the capacitors may be integrated into a monolithic device with the switch elements and/or may be external using discrete components. Similarly, at least some of the regulator circuit may in some examples be integrated with some or all of the charge pump in an integrated device.

Implementations of the approaches described above may be integrated into an integrated circuit that includes the switching transistors of the charge pump, either with discrete/off-chip capacitors or integrated capacitors. In other implementations, the controller that determines the switching frequency of the charge pump and/or the compensation circuit may be implemented in a different device than the charge pump. The controller can use application specific circuitry, a programmable processor/controller, or both. In the programmable case, the implementation may include software, stored in a tangible machined readable medium (e.g., ROM, etc.) that includes instructions for implementing the control procedures described above.

It is to be understood that the foregoing description is intended to illustrate and not to limit the scope of the invention, which is defined by the scope of the appended claims. Other embodiments are within the scope of the following claims.