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.
 
#52 · (Edited)
The gate driver board with protection features: UVLO at 18V and -5V, desaturation, and a high power isolation DC/DC. "Borrowed" from the SiC inverter project.

During the test phase all SiC mosfets will be driven by the this circuit to reduce the risk of "bang" incidents. And SiC does go "bang" at these power levels.

The new set of SiC mosfets *should* arrive today. Along with most of the gate driver parts.
 
#54 · (Edited)
Yeah, 390V is well below the 3kW module limit.

The gate driver board with almost all components soldered on.
First checks done. UVLO, LLC, hysteretic feedback, everything works fine. Only a few corrections (LT3082 pin mixup) and tweaks.

The 6 W LLC isolation transformer:
– Resonant inductance measured: app. 8.4 uH. Resonance at app. 160 kHz with a 100nF cap.
– LLC fixed frequency 200 kHz. Well above resonance.
– Q factor 0.8 at max load. Good value.
– Turns: primary 14 x 0,8 mm (to compensate for skin effect), secondary 2 x 29 x 0.4 mm.

About mosfet failures in LLC:
fairchildsemi.com/application-notes/AN/AN-9067.pdf
 
#55 · (Edited)
Solution for AN9067 issues:

With SiC diodes (no reverse recovery) (anti)parallel, the body diodes of the mosfets on the primary side of the 3kW LLC are bypassed, preventing high shoot-through currents in all cases, including startup.

The following points have to be addressed to get the power level above 750W:
1. PFC: the ICE2 controller power up (on/off) circuit, the software for it & desat protection.
2. LLC: SiC diodes on the primary side (LLC) (anti)parallel with the SiC mosfets
3. PFC & LLC: auxiliary supplies & custom UVLO protection @ SiC level (all)
4. The buck stage design
4. LLC: change the rheostat mode to potmeter for better accuracy and range.
5. Software: control LLC frequency and buck stage current.
 
#59 ·
Solution for AN9067 issues:

With SiC diodes (no reverse recovery) (anti)parallel, the body diodes of the mosfets on the primary side of the 3kW LLC are bypassed, preventing high shoot-through currents in all cases, including startup...
Generally speaking, you don't have problems with body diode conduction in the LLC converter if the switching frequency range is above resonance and loaded Q remains above 0.9-1.0.

If you want to put faster (e.g. - SiC) diodes in antiparallel with Si MOSFETs, anyway, note that both the higher forward drop of the SiC diode as well as the stray inductance between the internal body diode and the external diode will conspire to prevent the diversion of current to the external diode.
 
#56 ·
The hardware modification of the PFC controller on/off is a minor change to the existing LT3082 / FOD8320 circuit. Done in a few minutes.

PFC desat and UVLO

These automotive grade SOT23-6 components will be used to replace the "standard" UVLO reference and comparator circuits TL431 and LM2903:
http://www.ti.com/lit/ds/symlink/tps3808g01-q1.pdf
http://www.ti.com/lit/ds/symlink/tps3700-q1.pdf
Desat detection: ACPL-332J
Driver stays the same: IXYS IXDN609

All on a daughter board on the PFC PCB.
The software and daughter board can be ready in a few days, but the delivery of the TI parts may take two or three weeks.

LLC

The SiC diodes on the primary side were already part of the original design, so it's just a matter of putting them back on.

'Rheostat to potmeter' : minor change.
Protection: same daughter boards as the TI TPS 3700/3808 based PFC boards.

Auxiliary LLC supply: the RECOM RAC10/24SC is already in. Will be connected to the PFC 400V output.

BUCK STAGE FOR 3kW

The buck stage design and the software are going to take more time.
 
#57 · (Edited)
The high power gate driver board is up and running at 200kHz. At 24Vdc it takes 0.2A when loaded with two 3.3nF caps, simulating two 28A SiC mosfets in parallel.
The test result matches the engineering. Great.

HF AC PREHEATING
Thanks to GoElectric's topic "charger control lines"
https://www.diyelectriccar.com/forums/showthread.php/charger-control-lines-167466.html
I can still take HF AC pack preheating (in winter conditions) into account in the design of the buck stage.
Main difference: negative current threshold values for the hysteretic controller, allowing current to flow back from pack to charger (buffer cap).

Image: 1us/div, 5V/div, litlle bit of overshoot. The dampening can be tweaked.
 
#58 · (Edited)
Cracking on, because the 3kW has not been reached yet.
Next module, the gate driver board design with TPS3700 under & over voltage lockout and the other (SiC specific) features for the LLC & buck half bridges.

Bit early maybe, QUESTION : WIRELESS INTERFACE, what to use? smartpone/ GSM/ bluetooth ...

Ideas, suggestions?
 
#60 · (Edited)
Yes, but luckily the internal body diodes of SiC mosfets have even higher forward drop voltages.
So I don't think the inductances will come into play with my SiC mosfets.

During startup, the LLC runs at frequencies up to 600kHz. The application note describes an exceptional situation, specific for a half bridge LLC (resonant cap not charged yet), not full bridge, during startup.

Anyway, the exception had slipped my mind when I made the startup time longer AND removed the SiC diodes. Result: bang!

There is absolutely no problem for normal operation above resonance and at the right Q.

But, as I mentioned earlier, the Q factor is an issue at high loads.
The frequency has to be kept very close to resonance (narrow bandwidth).
So these circuits with well defined AC resistances are needed for LLC (rheostat/potmeter) frequency tuning:

Just those two circuits will do to tune all LLC circuits.
 
#61 ·
The ultra fast chargeable batteries have arrived:
http://insideevs.com/huawei-unveils-batteries-capable-ultra-fast-charging/

First application area is the mobile phone, but the batteries are suitable for EV use.
High energy density, 0 to 50% takes 5 minutes.

Even at 44kW, it takes more than five minutes to charge a 10kWh pack to 50%.

NEW DESIGN GOAL: 6.6 to 10kW per module

With the new charging requirement in mind, I've decided to set a new design goal. 6.6 - 10 kW will be the new default power level per module @200 to 425Vac in, 700Vdc max out.
400Vac in is for European 3phase 22kW+ line to line.
High isolation on the outputs, module outputs can be connected in parallel up to 1414Vpeak working voltage.

Development continues @3.5kW.
 
#63 · (Edited)
Sic does go bang, literally explosive, see pic.
Had to push it to really get to know the limits of SiC during LLC startup. Guess there will be more fireworks coming.:D

Sadly, not very soon, as there are more urgent matters that take priority.

And more pics: a rebuilt LLC power board (fresh SiC), a controller board with mods ( potmeter mode daughter board and more) and the new tune/ desat driver board.
 
#67 · (Edited)
CONTROLLER BOARD SELECTION: RASPBERRY PI2 (B)
(with an Arduino style SW lib flavour: wiringpi, wiringpi.com)

The board has 28 GPIO lines with two SPI interfaces, one has two select lines and the other three. And UART, I2C and more.

Wiringpi supports the low cost MCP23x08/17 I2C/SPI I/O expansion devices (US$ 3), so plenty of GPIO pins.

Raspberry Pi has a great Linux / open source software base: Raspbian (debian), ubuntu and more, connectivity (BT, LAN, USB, WiFi ..) and GUI stuff.

No real time OS, but all the critical timing is done on the dedicated module boards.
 
#68 ·
The RPi is a great choice. However, greedily I wish you could spend your valuable time working on the analog switching power supply details that few can do rather than figuring out how to get the pi to do what you already know how to do on the Arduino.

Hope you don't do what I found myself doing with the Pi. (Even though it was fun )
1. Learning linux.(well not so fun)
2. Learning Python because nearly all the cool stuff on the pi is written in it.
3. Learning that the book I've learned from uses Python3 when 4/5 the cool stuff is written in supposedly no longer updated Python2.
4. Yesterday learning the 'module' to do neat graphs, matplotlib hasn't been ported to Python 3.2 and will require 3.5 which isn't in the easy distribution for the pi.

So in other words there is much to learn, which you may like but won't get me a SiC charger any time soon:)
The Teensy 3.2 would have been my 1st choice, faster development, no OS, no IOT (Internet of things) to distract and faster at 72mHz than the pi at 700mHz.

Charge ON !
 
#69 ·
Hi Tony, good job with the development! No worries on the busted transistors - everyone working with these things has a solid collection of popcorn parts.
Tried to send you a PM here, but it seems you have them disabled. Can you send me a message with your email or something?
Thanks!
 
#70 · (Edited)
(@bicycleguy:
No, I'm not going to use Python. And Linux and C are my favorites for development, no worries there :) )

I have already ported the test Arduino test SW to the RPI. Pretty straightforward with wiringpi and SPI based I/O. First test run (SW only) went OK.

Unfortunately, there are more important matters that take priority, but I'm hopeful that I can do a full test run with all HW within a couple of weeks.

Edit: pictures of RPi SPI software test. Screen picture shows ADC values read through SPI and some lines of code.
Other picture: test setup with RPi and control board (SiC popcorn did not damage the board)
 
#71 ·
The next prototype version is on the way.

(RPi) SPI:
– To increase throughput on the SPI bus, the loose cables and flimsy connectors will be replaced with low cost flatcable parts.
The low speed (320kHz clock) LTC1598 ADCs will be replaced with smaller and cheaper 40MHz SPI clock LTC1863 ADCs.
– SPI GPIO expander MCP23S17 will be used for digital I/O, for instance for on/off control of LLC, PFC, BUCK, precharge relay, and fault feedback.
WiringPi supports up to eight expanders, so the maximum number of module sets will be eight.
– The TPL0501 digital potmeter is fast enough (25MHz) and stays for LLC frequency control.
Another TPL0501 will also be used to set the charge current in the buck output stage, all on one control board.

Popcorn related test results:
– Topology: Both topologies, halfbridge with SiC diodes antiparallel and full bridge (no SiC diodes), prevent high through shoot currents during startup.
No major difference in $$, full bridge needs more drivers, half bridge more cooling.
Half bridge does not perform well at high power levels (above ~1kW, poor efficiency, EMC), so it will be full bridge for all power levels.
– Drivers: the control boards are already adequately protected (protection diodes took the hit).
UVLO, desat and modified reset/ restart circuits will provide additional protection
for the SiC parts against high startup loads and fault induced stresses.

High amp gate driver:
The 200kHz output signal showed a little overshoot. Replacing the 3R3 gate resistor with a 6R8 creates a critically damped circuit.
The newest ACPL 339J driver board protects the PFC SiC mosfet.
With 10R the board is still faster than the gate output of the PFC controller.

Everything looks good, so the first high power test (~3kW) with RPi and extra SiC protection circuits is on the agenda.
 
#72 · (Edited)
First power test: PFC.
Again, with 2 230VAC heaters in series as load (750W, 1250W and 2000W selectable by switches). But this time all under full (RPi) control. ;)
The gate driver with protection features is clearly visible on top of the PFC module.
Borrowed from the 100kW+ SiC inverter, only one output used. :)
At the bottom of the picture is the mosfet power stage that switches the load on/off.
With another SiC gate driver board (first design) driving the high side of the half bridge.

Up to 2000W all went fine. Then the PFC shut own. No 400V DC @ 3.3KW but 325V with AC line ripple (effectively just above 2kW).
The only significant difference with earlier tests is the gate driver. To be continued.

A lot of preparation was necessary.
One example: measurement of the interrupt latency of the RPi in response to the gate driver fault signal: app. 150 microseconds,
which is well within the 1 millisecond fault hold time of the ACPL339J.

Other prep stuff: tuning board for the next test, get the LLC running at resonance.
Probably within a week, maybe two.
 
#73 ·
Ok, how stupid, it's a prototype. Just checked the design paper (math done about a year ago).
For 3.3kW two mosfets in parallel are needed. :eek: There is only one in this prototype version.
No wonder it turned into popcorn the first time at 3.3kW for a longer period of time. The gate driver board prevented a second serving. :)
The next version (summer recess) will have two.
 
#74 ·
I'd be interested in the lower power version. I'm still planning to proceed on my tractor EV projects which can use 1.5 or 2 kVA for maximum 2 HP motors. Let me know details on how to get one of the boards and some of the components, although I may already have many of them and I can order from Mouser or other distributors. Thanks. :)
 
#75 · (Edited)
Once I have ported the prototype versions (that is, with footprints for soldering by hand) to Kicad, I will put the gerber files online.
With BOM etc. Free for personal use.
Here is a picture of the start: creating the necessary libraries (thank you, mr. Rohrbacher!)
Edit: And pdf files of the buck dual high side gate driver prototype board (not yet tested).
 
#76 · (Edited)
#77 ·
Saw a Digikey add for this LLC controller chip.
https://www.fairchildsemi.com/datasheets/FA/FAN7688.pdf

The data sheet mentions some of the issues your addressing. I don't know if this might be useful but thought you'd like to see it. Lot of functions in a $3.00 part. The 'integral of the switch current' control and switching between pwm and frequency modulation depending on load seem cool. Also the non-ZCS and non-ZVS detection, (which I don't understand how they work)

Would any of this be useful to your design?
 
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