# Why does this 2-stage PWM charge pump have a very low efficiency?

I designed this 2-stage charge pump to get about 9V and 3mA (tolerances accepted) from a lithium-ion battery. The output works well and provides the requested voltage and current. I thought it is a very efficient thing till I measured the input current; it was about 45mA! I double checked everything but it seems correct. So for getting 27mW, I'm using ~170mW.

• PWM frequency = 10 KHz
• PWM duty cycle = 50%
• PWM top voltage = same as the battery (3.7 V in tests) and measures exactly half of the 3.7 as expected (50%)
• PWMs are from 2 different channels of a MCU but they are not phase-matched
• The capacitors in parallel are for better ESR and they are both MLCCs

What is the problem? Can I make it better to use it in a battery operated device?

• What voltage is PWM2? Have you simulated this? Commented Dec 15, 2023 at 13:19
• @tobalt The PWM2 specifications are the same as PWM1 (3.7V top) just from a different channel. No, I have not simulated it, just made a real prototype to check it. Commented Dec 15, 2023 at 13:22
• At your current, depending on the MCU, you might be able to drive the multiplier directly from the GPIOs, saving some transistors. You might also consider using a single stage x3 or x4 Dickson architecture, rather than your cascaded x2s. Is there a good reason not to use a readily available inductor-based boost upconverter? Commented Dec 15, 2023 at 13:35
• @Neil_UK I went for this instead of boost, because of the light load efficiency. I thought it's more efficient for just 2mA load. Commented Dec 15, 2023 at 13:40
• now your thoughts have been recalibrated, compare the actual efficiencies of various architectures for both capacitive multiplier and inductor boost, designed for the currents you want to use. Commented Dec 15, 2023 at 14:06

Here's a solution, BOM cost should be reasonable...

Rather than use two doublers, it adds the input voltage several times. It's simpler, because there's no need for level shifting or any active device. You either need two microcontroller pins (V1 and V3) to output opposing phase 100kHz square waves, or one micro pin and an inverter or any other inverting logic gate.

Output impedance is pretty high due to the 4 diodes in series, so it will output 12V when unloaded and 9V at 3mA. This problem is common with all diode based charge pumps, to have low impedance you'd need FET switches. You could regulate it by using the micro's ADC to check the output voltage.

Simulated efficiency is: 66% with 1N4148 or BAT54. But BAT54 (Schottky) gives higher output voltage, so there's more margin. 1N4148 with its higher Vf is borderline to get 9V output.

A tiny boost chip with an equally tiny inductor would no doubt give better efficiency and more features (like regulation). Whether it's worth it depends on the cost of energy (kWh from coin cells costs a lot!)

Q3 never fully turns off because the intermediate rail is too high for its control voltage. So you always get a resistive load current from this ~6V intermediate point to ground.

• So I believe I have to provide a PWM for that with higher voltage to turn it off. How should I do that? Commented Dec 15, 2023 at 13:28
• @TirdadSadriNejad One option would be to drive Q3 with a separate NPN transistor. But you will need another control signal for this (complementary to Q4's) Commented Dec 15, 2023 at 13:33

There are three issues that spring to mind.

1. Shoot-through: Have you looked carefully at whether Q1/Q3 turns off fully before Q2/Q4 turns on (and vice versa)? If not, you may be wasting significant power during switching. Dead time will help this hugely (and is the norm with setups like H-bridges)

2. Is the PWM2 control signal high enough voltage to turn off Q3? (as noted in tobalt's answer). Perhaps you need a pull up to the intermediate supply, and use PWM2 to drive a pulldown transistor?

3. Resistive losses: In that absence of inductors, capacitor charging / discharging will typically limit efficiencies.

Good summary in this app note.

However, perhaps worth looking at just chaining a pair of dedicated converters such as MAX1682?

(Image from above datasheet.)

These converters as cheap, reliable and have very low quiescent current consumption. Should be a really good match for your application.