john-joyce.co.uk

 Home | PhD

PhD Research

One of the more modern high-power semiconductor devices is the Insulated Gate Bipolar Transistor, or IGBT. It is displacing Gate Turn-Off (GTO) thyristors in increasinly powerful applications. For example, in railway traction, most trains built since the late 1990s use IGBT-based inverters [Heathrow Express, Northern Line, Juniper for example] whereas previous batches used GTO thyristors [Jubilee line, Networkers / class 365]. In conjunction with these modern semiconductors, multiphase induction motors have taken over from DC motors for reasons of lower maintenance and higher reliability. The older trains with DC motors used thyristors in later builds [from around the mid-1970s onwards; e.g. class 317], or tap-changers or series / parallel changes and resistors. What's got what is a useful reference for the equipment fitted to various types of trains.

High-Power IGBT Modules

My research focussed on the IGBT itself, and in particular the larger IGBT devices. Due to manufacturing limitations, IGBT chips can only be fabricated with current ratings of up to around 100 amps; beyond that, several chips are operated in parallel. A module can be bought with dozens of IGBT chips all connected together, with the assumption that all the chips will do the same thing at the same time.

Current Sharing and Redistribution

It turns out that the parallel chips do not always share current evenly, and in particular, during switching. At turn-off, some chips turn off faster than others, and so current concentrates in the slowest ones until they turn off too. We developed a current measurement system which can be placed inside the IGBT module, around the internal connections, to investigate this issue.

installation
in a typical module
A typical current measurement installation in an older IGBT module

Above can be seen a typical installation of the current measurement system in an older IGBT module. (click for a bigger picture) This particular module has four chips, connected together in two pairs, and is intended to be used as one 'leg' of an inverter.

The current probe developed by Bernard Stark and myself is shown around the internal connections. A detailed description of its theory and operation can be found in our paper at PESC 1997; in summary, a current in a conductor produces a magnetic field around that conductor proportional to the current. Changes in this magnetic field induce a voltage across any other conductors nearby, so by placing a coil nearby and integrating its output, we can recover the original current waveform.

the field
probe itself
One of the field probes we developed. Its small size allows its use in very confined spaces.

If this sounds like a Rogowski coil to you, it is, more-or-less. It is subtly different in that it consists of two parallel coils rather than a complete toroid. It was designed to be retro-fitted into existing complete devices, so a toroid was not practical at this scale.

Current redistribution
Current redistribution at turn-off. The two collector current traces show current moving from chip 2 to chip 1. As can be seen above, IGBT modules don't always behave as they should. Similar results were shown with thermal imaging techniques.

Initial investigations centred on practical measurements. Some work with different gate topologies turned up potential improvements to module performance. The consequences of using sophisticated gate drives were also investigated and the results published in PEVD 1998. Selecting IGBT chips for assembly to optimise performance was investigated. However, this wasn't very controllable.

The next line of attack was simulation of parallel chips; much more carefully controlled variations could be introduced into the chips and the effects investigated. A fair degree of success was achieved in this way, although the effects predicted were larger than those seen in reality. This was published in ISPSD 1999.

Oscillations
Current oscillations between two chips

An additional angle was the presence of differential oscillations between chips under certain switching conditions - in this case, the module was operating at a fraction of its rated current and voltage. A small-signal model for the IGBT was developed and analysed mathematically to predict its stability against such oscillations. It was found that such stability is heavily dependent on both the IGBT and the design of the module. Some of these issues are documented in the EPE 1999 paper.

Publications


Last updated 26 Dec 2008 23:01 by John Joyce - email jcjoyce@iee.org