Nov 2, 2016 | By:

Posted in: , , ,

As a follow up to my last blog about the pros and cons of using discrete IGBTs or IGBT modules in different inverter designs, one essential question for design engineers is how a discrete IGBT performs when it’s used in parallel.

In some designs, the current sharing among paralleled devices might not be well balanced, with some parts carrying more current than others. This creates extra power loss and thus a hot spot on these parts. Long term operation under such conditions can significantly deteriorate a device’s reliability and shorten its lifetime. In extreme cases, a heavy-loaded device could be damaged immediately.

Sometimes unbalanced current sharing is due to non-optimized circuit design, especially the external circuit around a discrete IGBT and their drivers. Sometimes it can also come from the IGBT itself. Lot-to-lot variance and variance within the same lot can make a device perform differently than its peers. For example, a device with a smaller Vge_th will turn-on earlier and turn-off later, thus a large portion of system current will pass it during switching. Similarly, a device with high Vce_sat will generate more heat and raise Tj faster.

Design engineers can improve the circuit design if the unbalance has an external cause. But there are only few things they can do if the unbalance is from the IGBT itself. Therefore, to get the best performance when using multiple discrete IGBTs per switch it is vital for design engineers to know how a device is made and how it will perform when used in parallel. That’s where a Cloud Map comes in.

As an IGBT supplier, ON Semiconductor not only defines IGBTs per customer needs at the design stage, but also implements well-controlled manufacturing processes from the fab, through assembly, to the final test. Our commitment to excellence also includes using a Cloud Map to correlate Vce_sat and Vge_th data for selected manufacturing lots. The Cloud Map includes two important parameters for determining an IGBT’s parallel performance, Vge_th and Vce_sat – the narrower the Vge_th distribution, the more uniform IGBT switching; the closer the Vce_sat among paralleled IGBTs, the less conduction loss variance. Such a Cloud Map can help design engineers choose the best device for parallel usage, or conversely, it can help them define the system tolerance in parallel operation when an IGBT has already been chosen. In cases where both the system design and the IGBT have already been chosen and only minor modifications are possible, a Cloud Map can help design engineers and suppliers decide if binning is needed and what’s the binning criterion. Moreover, by comparing the Cloud Map at different times, design engineers can gauge how stable the production process is and whether binning criteria needs to be revised.

Here we’ll use Cloud Map as a starting point to evaluate the parallel performance of discrete a IGBT – a Cloud Map of 6,700 units of FGY120T65SPD_F085 IGBTs from 10 production lots is shown below. The ΔVce_sat at 120A and ΔVge_th at 120mA are smaller than 0.15V and 0.5V, respectively.

screen-shot-2016-11-02-at-3-20-04-pm

In order to verify how a Cloud Map can help judge device performance when used in parallel, we also conducted a current sharing test similar to the one below, where a half-bridge circuit is built with multiple FGY120T65SPD_F085 IGBTs in parallel per switch. This circuit is carefully designed in order to minimize the external circuit impact on the current sharing.

screen-shot-2016-11-02-at-3-20-25-pm

By randomly choosing five lots of FGY120T65SPD_F085 IGBTs (each with 12 units) from the 6,700 unit production samples mentioned above, 60 combinations of two devices in parallel were tested and the current passing each IGBT was measured separately at maximum output (around 120A current for each IGBT).

screen-shot-2016-11-02-at-4-16-04-pm

Test results showed a very good current sharing performance – the average current difference was around 0.4 while the maximum value was around 2%, which is less than 3A. Such a small current difference was recognized and appreciated by our customers as it helped them to use the FGY120T65SPD_F085 IGBTs in parallel.

screen-shot-2016-11-02-at-3-20-42-pm

From the discussion above, we can see that a Cloud Map can help design engineers in multiple ways including a parallel performance study of a discrete IGBT. The effectiveness of a Cloud Map is an excellent complement to a current sharing test.

Visit the HEV page to learn more about our HEV offerings.

This post is also available in: Chinese (Simplified)