DIY Electric Car Forums banner

SiC LLC modular charger design

75K views 272 replies 20 participants last post by  Tony Bogs 
#1 · (Edited)
Inspired by PStechPaul's idea of a modular charger and the availability of SiC and LLC devices,
I've decided to go ahead with a SiC / LLC based modular charger design / DIY homebrew build.

Design choices:

  • A SiC PFC CCM boost input stage, 390V DC out.
  • 3350 W out, (230V, 16A AC in) per module.
  • LLC transformer topology for isolation and additional boost up to 580V (for HV battery packs). FIXED FREQUENCY. Boost: by transformer turns ratio only.
  • CC mode only, buck hysteresis current control
The measurement of the LLC inductance (36.9uH) is an important step in the design proces.
 
#29 ·
Next is control of the LLC.
There's no off the shelf dedicated controller that can hold the LLC at series resonance AFAIK. So MCU control and monitoring is needed.

The LLC will be operated in open loop. At series resonance, where the efficiency is at maximum and the DC gain is highly independent of the load.
So a complicated analysis over a wide frequency range is not necessary.
Still interested in the full DSP analysis for a closed loop approach? This can be skipped: http://ww1.microchip.com/downloads/en/AppNotes/01477a.pdf
The choice for an Arduino and SPI ADC is easily made.
With SPI ADC's only a few digital lines need the full 10 mm isolation.
Regarding the software side:
The Arduino IDE makes it easy to get started, the AVR Arduino's are supported by the open source community and the SPI functions are immediately available in the IDE top layer.
Great for a homebrew project on a DIY forum.

As it stands now, Linear LTC1598 SPI 12bit, 8 channel ADC devices will be used. There's no need for high speed. All signals of interest are DC or LF.
There are two isolated circuits to be monitored: the input side (mains in, PFC, LLC primary side) and the battery side (LLC secondary, buck regulator, battery).

Short description of the MCU LLC control:
At timer tick intervals the MCU determines the DC gain of the LLC converter at frequencies close to resonance. Series resonance is just above the frequency where the gain is at maximum.
 
#30 ·
The Fairchildsemi FAN7631 https://www.fairchildsemi.com/datasheets/FA/FAN7631.pdf will control the LLC half bridge.
It handles soft start, bootstrap level shift, overcurrent and under/overvoltage protection, dead time insertion, .....
The goal is to keep the software effort minimal.
In stead of the usual optocoupler feedback circuit the LLC frequency will be adjusted using a SPI rheostat: the TI tpl0501 http://www.ti.com/lit/ds/slis136a/slis136a.pdf

Tranformer isolation.
For the isolation the large airgap and high leakage properties of the LLC will be used.
The two halfs of the ETD59 core will be glued together with a copper free 1 mm FR4 epoxy board between them.
No coil former will be used. Only FR4 spacers and high temp polyester film.
 
#31 ·
I see - so you are using the SPI rheostat to control it via a microcontroller rather than using a feedback network. Interesting.

This will only be for voltage feedback though? Or are you adjusting the CC value using the voltage control (makes sense).

Looking at the topology it has Rcs (current sense resistor) for current limiting. This would probably be more for device protection than constant current limiting.
 
#32 ·
Thank you for asking.

Capacitive current sensing (divider, less power loss) is implemented, but a ground sense resistor is also possible.
Current sensing is indeed for protection, not control.

LLC with feedback voltage control (optocoupler) in a charger is very challenging, because the AC impedance of a battery is very low.
The DC gain of a LLC is AC load dependent, with one exception: at series resonance, where the tank impedance is zero.
That property is used in the charger design, since the input voltage is regulated (by the PFC, at 400V).
Although the dynamic load regulation of the PFC regulator is very poor, that's not a problem in a charger,
because the load varies very slowly in time.

A hysteric controlled buck regulator regulates the charge current. As “seen” by the LLC, the buck regulator has a much higher impedance than the battery, so some control is possible ( i.e. hold LLC at resonance).
The DC gain at resonance can be calculated. In this case 1.33. Rock solid with almost zero impedance. Almost, cause there are losses in the mosfet Rdc, winding resistance …

With the SPI rheostat the microcontroller can hold the oscillator of the FAN7631 controller at the series resonance frequency, where the DC gain is 1.33. So one could call it a form of (hysteretic?) gain control.

Purpose of the LLC in the charger: provide voltage boost with high isolation and a very low capacitive coupling.

Another big LLC advantage: even below resonant frequency, down to the frequency, where the DC gain peaks, ZVS is maintained for the halfbridge mosfets. Thus, losses can be compensated (at least partially). http://www.diyelectriccar.com/forums/showthread.php?t=162082
 
#33 ·
The SPI control and monitoring board layout is almost done. A few minor tweaks still to do.
Control: LLC on/off and LLC frequency (SPI rheostat)
Monitoring: temperatures (eight), PFC input current, charging current (sense R, single module), battery pack voltage,
LLC in- and output voltages, charging current (hall sensor, all modules combined) and brown out detect PFC input voltage (bridge out).
One spare volt input.
Assembly is scheduled for saturday.
 
#35 ·
No schematic yet (still pre 0-series), but I'll add references to the silk when boards are to be ordered pre-assembled (SMT part). To do so, I'll have to use the design package of the selected supplier.
The black stuff is indeed the silk layer in the 'prototyping' design software. I'm using it to create a printable overview of what part goes where for manual assembly.
 
#36 · (Edited)
The SPI control and monitoring board is ready for the first (basic) tests.
The SPI rheostat and ADC's will be soldered on after the succesful completion of the first tests.
I'll connect the board to an Arduino Due (with 32bit ARM processor), but since it is a SPI board (MCU independent, modular) many other development boards are suitable.
Maybe I'll also try a STM32 with the GNU toolchain on a Linux system.

Several fixes were needed based on the first tests and checks:

  • LM1117I (cheap old LDO) didn't work, replaced it with a new one, no result, tried LT3082: does work.
  • Added parts: pull down resistor, zener protection for LT1461 inputs, opamp bias resistors (in case input is not connected), rerouting layout errors with pieces of wire wrap.
 
#37 · (Edited)
The LLC is up and running. At low input voltage (24VDC), with a light load (<10W light bulbs) and at a frequency just below resonant.
No ZVS under these conditions, there is ringing at the LLC transformer/SiC junction.
Measured voltages: input 24,43V, output 32,42V. DC gain is 1,327. Very close to the calculated (design) value of 1.33.

Everything is looking OK for further tests with higher loads, voltages and frequency.

24VDC in, 40 Ohm load:
The reflected AC resistance is ~ 5. Half of the Rac ~= -10 at full 3300W load with buck end stage.
Measurements show very little ringing at frequencies above resonant.
The LLC will run at about 112kHz.
With the SPI ADCs and rheostat on the board it's time for software.

OK, the Arduino Due controls the LLC frequency. Wrote a small program that varies the SPI rheostat resistance slowly min to max and back.
 
#38 · (Edited)
The SPI board works with an Arduino Due: set the rheostat, read ADC and select ADC inputs. Picture of the test circuit is in the attachment.
More software needs to be written to hold the LLC at resonance. To test the software reliable interfaces / connections between Due, SPI board and the LLC power module are needed.
Probably takes about a week to write the software and get the hardware ready.
 
#39 · (Edited)
Scope still image and gzipped mp4 video (rheostat control) of the voltage at the junction of the SiC devices and LLC transformer (primary tank input).
10V per division, 0.2 usec/div timebase. LLC input voltage app. 45V.
Clearly visible, there is still hard switching at this low input voltage, but as the frequency increases, ZVS (of the high SiC mosfet) is almost reached.
Jpeg: at 400 nsec the low SiC mosfet turns off, at 600nsec the high SiC mosfet turns on.
The ringing is normal under hard switching conditions.

In order to ensure ZVS under normal operating conditions, the dead time has to be increased.
The relationship between primary inductance, bridge capacitance, operating frequency can be found in: http://www.infineon.com/dgdl/Design...1.pdf?fileId=db3a304330f68606013103ebd94f3e98
The design calculations are based on a transformer with a larger airgap (special construction, lower primary inductance with a 0,5mm FR4 PCB as isolation barrier).
The tests are done with a standard ETD49 coil former.

At 90V and 325V input and with increased dead time still no ZVS.
There is conflicting information in application notes about secondary side ZCS above resonance.
Earlier designs always had schottky diodes on the secondary side, so that's the first possible cause to investigate.

Yeah, recovery losses (second image): top trace is primary voltage, lower trace = secondary voltage. Secondary lags by about 400 to 500 nsec, which is the recovery time for the secondary Si diodes at low reverse recovery current.
Conclusion: go full SiC (no recovery losses).

Found the pdf of the Cree 8kW reference design:
www.cree.com/~/media/Files/Cree/Power/Articles%20and%20Papers/White%20Paper%20Highly%20Efficient%20and%20Compact%20ZVS%20Resonant%20Full.pdf
 
#41 · (Edited)
YES, ZVS.
Turns out I was running the LLC at 125-135 kHz in stead of 90-100kHz.
Made a mistake when I measured the resonant inductance and forgot to change the timing on the LLC board.
Maybe I'm getting too old for this stuff.

Never mind, the 4 diodes were only $12.
So let's crank up the power.
 
#43 · (Edited)
The high power ferrite set is in. Running at 250kHz in a full bridge LLC configuration, it *should* be able to transfer 10kW. The pen is about 5".

Test result, confirmed by Cree's description of the 8kW conventional LLC: certainly no need for SiC diodes on the primary side. Reduced losses.
SiC diodes on secondary side stay: reduction of dead time.
 
#44 · (Edited)
Scope image of ZVS (zero volt switching) of bottom SiC mosfet.
Top trace: gate-voltage voltage, 10V/div
Bottom trace: Drain-source voltage, 100V/div
Time base : 1us/div
LLC load: two 75W light bulbs in series

The drain-source is already zero for about 250 nsec when the gate signal rises to turn the mosfet on.
 
#46 · (Edited)
No, didn't go all that well.
Unfortunately, I had to switch to another computer system last weekend.
The eight year old 2-core gave up.

And the PFC SiC mosfet died. Not SiC related, but the PFC controller is sensitive to the rise and fall times of the auxiliary power supply.
So far I've always used a simple “plug a wire end in a connector” method for PFC turn on/off.
This time the controller didn't reset properly.

The picture shows the test rig. The connector is above the PFC elcaps.

3kW test postponed to next weekend without PFC boost: 310V LLC input.
I'll order some fresh mosfets. Delivery may take three weeks.:(

The connector issue is easy to solve via the Arduino Due. For the 10kW+ design, I'll use drivers with desat and UVLO protection.
 
#47 · (Edited)
With the convection heaters at 750W setting (total 1500W), the LLC didn't start.
The overcurrent protection keeps kicking in.
It also kicked in a few times with a 150W load during startup. 150W light bulbs = 1500W when cold at startup.
So it seems to be a startup issue.
The output cap probably isn't fully charged when the LLC reaches the high power frequency range.
Next try: turn the load on after startup when the output cap is fully charged.

Time for the last one of the 3500W modules: the buck stage.
 
#48 · (Edited)
The SiC 18 to 19V UVLO circuit for the GATE DRIVER SUPPLY. Not yet tested.
SiC needs higher UVLO thresholds than the usual IGBT values (11 to 14V).
If the driver positive supply voltage is too low, the voltage regulator for the ACPL-339J optocoupler is shut down. The 339J then generates the UVLO fault feedback.
http://www.mouser.com/ds/2/38/AV02-3784EN_DS_ACPL-339J_2015-04-27-480718.pdf

The ACPL-339J will drive the high side SiC in the buck stage. Improved UVLO (with IXYS 609) in second image.
The 339J has a high thermal resistance and a low maximum junction temperature (125C). It can't drive the intermediate mosfets directly at high frequencies.
The modified PMOS drive circuit is in the third image. Similar circuit for NMOS is not shown.
 
#49 · (Edited)
I've modified my half bridge inverter that I've used earlier, so it can be used as a buck stage. The frequency is set at 10kHz with duty cycle 0%, ~50% and 100% selectable with a simple jumper.

The first (functional) check at 50% duty cycle, using light bulbs as load, went OK.
Test result with heaters as load, 1500W, 50% buck, continous from startup: same startup issues as earlier, overcurrent protection keeps kicking in.
Test was repeated with buck at 0% during LLC startup, then switch buck to 50%: OK. LLC output voltage at 350V.
Next test was at 100% (1500W): shutdown.
So the LLC is at 750W now.

Done some more measurements. Seems to be the 20V power supply for the LLC controller. Can't supply enough mA.
 
#50 · (Edited)
Saw a video yesterday about a very simple charger based on a rectifier and (hugh) capacitors as current limiters.
Non isolated and a power factor as low as 50% :eek: . But it is simple and it works.
It remindend me not to overengineer things.
Didn't work for the power supply for the LLC controller though.
It is not possible to use the same P/S for PFC and LLC.
There's a high current path in the LLC P/S return. Solution: an isolated converter for the LLC controller.
 
#51 · (Edited)
The CCM current mode hysteretic buck controller.
To keep it simple: no hardware SiC desat/UVLO/overcurrent protection.
Maximum output current app. 50A peak. Max output voltage: app. 500V. Min output voltage: app. 48V.

Protection features (left them out for clarity):
- Rsense can be used to detect overcurrent with a LM2903 comparator circuit, bypassing the Hall sensor
- UVLO protection on the high side is half of the 339J UVLO circuit I posted earlier: just the 2903 and 431 sections.
- Desat is not needed here, because the low side of the power stage is passive (diodes). Short circuit is very unlikely.
The 10kW version will have a Si mosfet in parallel with the SiC diodes to reduce loss. With desat protection.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top