EEVblog Electronics Community Forum
General => General Technical Chat => Topic started by: tom66 on February 26, 2014, 11:53:15 am
-
Engineering pr0nography with thanks to "Marco2228" of Tesla Motors Club -- here's what's inside the power pack for a 300hp/200kW performance electric sportscar. He's the first to attempt a repair of such a unit after damage occurred (I doubt anyone would be tearing down a working one...!)
I'm quite surprised that they used discrete IGBTs rather than dedicated hybrid modules, but it looks to be a cost decision.
Charging input / contactors (240V @ 70A)
http://www.teslamotorsclub.com/attachment.php?attachmentid=44149&d=1393359626 (http://www.teslamotorsclub.com/attachment.php?attachmentid=44149&d=1393359626)
Two phases ("Megapole logic") of inverter drive - identical to each other
http://www.teslamotorsclub.com/attachment.php?attachmentid=44154&d=1393359673 (http://www.teslamotorsclub.com/attachment.php?attachmentid=44154&d=1393359673)
High performance DSP for traction control, safety interlocks, ADC/DAC and motor drive plus fan controller (far right)
http://www.teslamotorsclub.com/attachment.php?attachmentid=44151&d=1393359647 (http://www.teslamotorsclub.com/attachment.php?attachmentid=44151&d=1393359647)
One IGBT "module" (a "Megapole") - 110kW IGBT block, 350VDC @ 650A peak (~310Arms)
http://www.teslamotorsclub.com/attachment.php?attachmentid=44150&d=1393359635 (http://www.teslamotorsclub.com/attachment.php?attachmentid=44150&d=1393359635)
Thread:
http://www.teslamotorsclub.com/showthread.php/20709-34-Salvage-auction/page11 (http://www.teslamotorsclub.com/showthread.php/20709-34-Salvage-auction/page11)
.
-
Aw doesn't seem to mention what IGBT part they're using.
That's pretty awesome though, 28 IGBT's per phase.
Some serious power switching there.
-
Very interesting find!
28 TO-247 IGBTs per phase is indeed an interesting design choice. Considering the thing can put out 200kW at least for some time it's very compact.
-
That is pretty awesome. Corp666 was right, pretty soon Summit Racing will be selling Souped-Up boards with 96 IGBT's/phase.
Wonder when the local track's Supercharger will be installed?
-
I was expecting a large FPGA or two to handle the low level modulation and control. (There might be, just not pictured.) Hard to tell for sure, but it might also be a multi level inverter with all those transistors and capacitors per phase.
Meanwhile, Toyota directly mounts IGBTs dies into modules and uses custom ASICs, but they're also dealing with much higher volumes.
-
Model S uses Ixys IGBT's. there is a reason for not using modules : individual monitoring and control. These guys are not stupid ...
-
I'm going to do that myself one of these days, that is, make an array with the transistors on the backside so the heatsink is clamped by screws through the board.
Discrete transistors are so damn cheap, it almost doesn't make sense to use modules. The main difference is convenience and serviceability, I think. For sure, if you're doing something at high frequencies, there is no module in existence that can go fast -- think the highest I've seen is a Microsemi/APT MOSFET half-bridge only rated for 200kHz resonant (compare to a fast IGBT bridge that wheezes out around 50kHz, or discrete MOSFETs that'll do MHz these days). It appears they made no effort whatsoever to reduce lead inductance, which is shameful.
Hmm, also curious what isolation they use. All those heatsink pads are at different potentials (B+ and AC, respectively). Wonder if it's hardcoat anodized...
Tim
-
Discrete transistors are so damn cheap, it almost doesn't make sense to use modules. The main difference is convenience and serviceability, I think. For sure, if you're doing something at high frequencies, there is no module in existence that can go fast -- think the highest I've seen is a Microsemi/APT MOSFET half-bridge only rated for 200kHz resonant (compare to a fast IGBT bridge that wheezes out around 50kHz, or discrete MOSFETs that'll do MHz these days). It appears they made no effort whatsoever to reduce lead inductance, which is shameful.
I'm surprised just how many modern digital audio amplifiers are using through hole modules when surface mount discrete MOSFETs and chips are superior at high switching frequencies. Of course, they deal with small amounts of power as far as power electronics go.
-
Hmm, also curious what isolation they use. All those heatsink pads are at different potentials (B+ and AC, respectively). Wonder if it's hardcoat anodized...
Tim
if the transistor is built on fully depleted silicon no insulator is required. That thermal slug has no electric connection
-
Building my own electric car and making the speed control unit is something i would find quite fun.
Probably wouldn't do regen braking to keep things simple but testing out different control systems would be interesting.
Not sure about the safety aspect though, I wouldn't want some code error to cause fullspeed lockup while driving with other cars nearby.
In NZ all electrics cars are required to have a control for the driver that mechanically disconnects the batteries so E-stop would always be possible in an emergency.
Multi kW speed controller design is probably a lot more complex than i think it is but it still sounds like fun.
DC pwm and a brushed motor would be the easiest
-
For safety interlock, the simplest system would be two identical processors cross-checking each other, with each having an independent accelerator input from the pedal. The gate drive signals each generate could be AND-ed for the final drive signals, and an XOR-circuit could be generating an interrupt indicating a processor exception which would immediately result in the battery contactors opening if the disagreement time exceeds some pre-set limit. In addition, applying the physical brake hard-down should command the inverter to stop delivering power.
In the Roadster PEM it looks like the Xilinx FPGA/CPLD near the phase connector might be doing some cross checking. It's also possible the checks are done outside the PEM, maybe in the vehicle management module, and that the PEM is fairly "dumb" - just taking a torque input command and producing an output. But the TI DSP seems like a bit much if it's just doing that. There's definitely some traction control management going on there as there's some tech docs talking about how the original "analog PEM" (presumably full of op-amps with no DSP...) without traction control being terrible at handling.
-
if the transistor is built on fully depleted silicon no insulator is required. That thermal slug has no electric connection
I don't think a device like that has ever existed...? It certainly wouldn't work with a minority carrier device.
All modern switching MOSFETs and IGBTs are either some variant of VDMOS (the die backside is doped for the drain/collector connection, built around a lightly doped wafer, which becomes the drift region) or superjunction (the drift region is alternating N/P columns). In either case, the backside is always a terminal, and in TO-247 packages, almost always physically connected to the center pin (cut and formed from a single piece of tin plated copper).
There are isolated tab devices, usually using DBC (an aluminum nitride type ceramic insulator with copper plates direct-bonded to both sides). I suppose they could've used these, but I doubt the cost would be justified. (They probably did use hardface anodize... it's cheap, and easy to apply to a lightweight aluminum heatsink, which probably needs anodizing anyway for corrosion protection.)
Tim
-
if the transistor is built on fully depleted silicon no insulator is required. That thermal slug has no electric connection
I don't think a device like that has ever existed...? It certainly wouldn't work with a minority carrier device.
All modern switching MOSFETs and IGBTs are either some variant of VDMOS (the die backside is doped for the drain/collector connection, built around a lightly doped wafer, which becomes the drift region) or superjunction (the drift region is alternating N/P columns). In either case, the backside is always a terminal, and in TO-247 packages, almost always physically connected to the center pin (cut and formed from a single piece of tin plated copper).
There are isolated tab devices, usually using DBC (an aluminum nitride type ceramic insulator with copper plates direct-bonded to both sides). I suppose they could've used these, but I doubt the cost would be justified. (They probably did use hardface anodize... it's cheap, and easy to apply to a lightweight aluminum heatsink, which probably needs anodizing anyway for corrosion protection.)
Tim
There are loads of devices with insulated tabs. First one I found with 10s of searching: https://www.fairchildsemi.com/ds/HG/HGTG30N60B3.pdf (https://www.fairchildsemi.com/ds/HG/HGTG30N60B3.pdf)
-
Would the isolated tab not make it harder to keep the IGBT cool?
-
Does not look like it does, seeing as max thermal resistance is 0.6 C/W. Even copper is around that figure, so the internal isolation must be pretty good.
-
I think I recall them saying the vehicle is 86% battery-to-wheel efficiency
Probably breaks down around 95% inverter efficiency, 95% motor and 95% battery.
This means at peak power the inverter is dissipating over 10kW, or each device on average around 120W... (assuming IGBTs are the biggest lost, sounds about right)
So they'd better get the cooling right!
-
There are loads of devices with insulated tabs. First one I found with 10s of searching: https://www.fairchildsemi.com/ds/HG/HGTG30N60B3.pdf (https://www.fairchildsemi.com/ds/HG/HGTG30N60B3.pdf)
Did you paste the wrong link by accident..?
Here's an example,
http://www.digikey.com/product-detail/en/IXGR48N60C3D1/IXGR48N60C3D1-ND/2117159 (http://www.digikey.com/product-detail/en/IXGR48N60C3D1/IXGR48N60C3D1-ND/2117159)
but it's a little more expensive (though the ratings aren't quite equal, either).
Tim
-
Here's an example,
http://www.digikey.com/product-detail/en/IXGR48N60C3D1/IXGR48N60C3D1-ND/2117159 (http://www.digikey.com/product-detail/en/IXGR48N60C3D1/IXGR48N60C3D1-ND/2117159)
but it's a little more expensive (though the ratings aren't quite equal, either).
Tim
That's an interesting part. Funny enough, a big FET like the FDH44N50 costs about the same, has very similar ratings, and would dissipate about the same amount of power at 55A (about 160 Watts):
http://www.digikey.com/product-search/en?WT.z_header=search_go&lang=en&site=us&keywords=fdh44n50&x=0&y=0&formaction=on (http://www.digikey.com/product-search/en?WT.z_header=search_go&lang=en&site=us&keywords=fdh44n50&x=0&y=0&formaction=on)
But those aren't available with an insulated tab.
-
It seems that they are using these from a little hint by Marco
http://datasheet.octopart.com/IXGX120N60B-IXYS-datasheet-11760468.pdf (http://datasheet.octopart.com/IXGX120N60B-IXYS-datasheet-11760468.pdf)
about $20 in onesies, $10 in high quantities.
Driven by a TC4451 IGBT driver, which is a Microchip part. A 12 amp IGBT driver; I've never seen one that high before, I thought most of them topped out around 4 to 5 amps. Microchip in a Tesla -- a pretty big endorsement of Microchip's analog/power division; then again, I have often found them to be the only manufacturer of some parts I needed. Perhaps this was the same in Tesla's case.
Tab is not isolated.
-
The only other high capacity gate drivers I know of, offhand, are the IXDD614 series (414, 614, different packages, invertering/non, etc...), which go up to 30A or so. There are probably others, I just haven't looked in forever.
Micrel often second-sources chips like that... or Microchip is the second source, I don't know. :P Offhand, Digikey lists the MIC4451 (non stock), though they don't have an exposed-pad type (which is probably a good idea for something like this?), which Microchip does.
Tim
-
Would the isolated tab not make it harder to keep the IGBT cool?
no. it is not a problem. there is still a heat transfer inside. either fully depleted silicon or a slice of aluminum oxide.
we can grow depleted silicon on the backside of the wafer. that acts as barrier.
Tesla uses a 3 phase 4 pole induction motor. Other EV's are all permanent magnet BLDC constructions.
-
Well it's irrelevant because they use a nonisolated tab so I figure they just isolate them externally, not sure how though.
-
Would the isolated tab not make it harder to keep the IGBT cool?
no. it is not a problem. there is still a heat transfer inside. either fully depleted silicon or a slice of aluminum oxide.
we can grow depleted silicon on the backside of the wafer. that acts as barrier.
Isolated devices are built with traditional thick film ceramic construction techniques. An insulator (these days, probably aluminum nitride or oxynitride, but oxide is the classic material) with metal plated on both sides. The top side is etched (chemically or by laser) to make electrical connections as needed, especially handy if co-pack'd with a diode.
I know this is typical of SOT-227 packages and other modules. It's probably typical of the isolated types too, but I haven't broken any open yet.
Intrinsic silicon is not used as an insulator, never has been(?), and never will be. It is impossible in general to use for that purpose.
Tim
-
It seems that they are using these from a little hint by Marco
http://datasheet.octopart.com/IXGX120N60B-IXYS-datasheet-11760468.pdf (http://datasheet.octopart.com/IXGX120N60B-IXYS-datasheet-11760468.pdf)
about $20 in onesies, $10 in high quantities.
Excellent, i know what IGBT i'll be using in my solid state tesla coil upgrade :D.
-
It seems that they are using these from a little hint by Marco
http://datasheet.octopart.com/IXGX120N60B-IXYS-datasheet-11760468.pdf (http://datasheet.octopart.com/IXGX120N60B-IXYS-datasheet-11760468.pdf)
about $20 in onesies, $10 in high quantities.
Excellent, i know what IGBT i'll be using in my solid state tesla coil upgrade :D.
I've gotta say, I'd love to see a 10HP SSTC :o :o >:D
Tim
-
great thread guys.
-
There are loads of devices with insulated tabs. First one I found with 10s of searching: https://www.fairchildsemi.com/ds/HG/HGTG30N60B3.pdf (https://www.fairchildsemi.com/ds/HG/HGTG30N60B3.pdf)
Did you paste the wrong link by accident..?
Here's an example,
http://www.digikey.com/product-detail/en/IXGR48N60C3D1/IXGR48N60C3D1-ND/2117159 (http://www.digikey.com/product-detail/en/IXGR48N60C3D1/IXGR48N60C3D1-ND/2117159)
but it's a little more expensive (though the ratings aren't quite equal, either).
Tim
How come? There is no mention that the tab is connected to anything. In my experience that is the case then. For a counter example this has a clear marking that the tab/flange is not isolated: https://www.fairchildsemi.com/ds/FG/FGH75N60UF.pdf (https://www.fairchildsemi.com/ds/FG/FGH75N60UF.pdf)
-
Lack of information is not positive proof to the contrary of that information... ;)
It's doubly dangerous, because 1. it doesn't specify that the pad is connected (however, I would assume by default that it is, particularly because a standard TO-247 package is specified), and 2. if it IS isolated, the isolation standoff is not specified -- note in the IXYS 'ISOPLUS247' case, it's rated for 2500Vrms. So even if it were isolated, you have absolutely nothing to rely on.
While you can make whatever hand-waving you wish (maybe even buy some and inspect or test them yourself), without #2, you are guaranteed nothing. What's more, the manufacturer is free to change the part under that constraint (or the lack thereof) as they see fit. (Though most datasheets include that nasty little phrase, "subject to change without notice", or something to that effect, making that kind of moot anyway...)
Tim
-
Intrinsic silicon is not used as an insulator, never has been(?), and never will be. It is impossible in general to use for that purpose.
You may want to update your sources.... We have a process on fully depleted silicon to make very low power (as in quiescent currents) devices. I don't know if this process is used to make power transistors. But it does use a layer if fully depleted silicon because of the electric isolation. FD-SOI fully depleted silicon on insulator. We already have a number of devices i. This process. So does intel and a vunch of others.