Converter-Level Integration for Wide Band-Gap Automotive Power Electronics

Existing approaches in power electronics have served the automotive sector well, but with the advent of electric vehicles, silicon based devices and converter topologies are reaching their performance limitations. The wide range of requirements in the automotive sector is driving demand for higher temperature materials, higher switching frequencies, improved reliability and more power dense solutions. In addition, new, promising technologies such as wide band-gap (WBG) semiconductors, need scaling up to meet the demands of the automotive sector, which creates both technical and manufacturing challenges.

The packaging of semiconductor devices is critical to managing their mechanical stability in addition to their thermal and electrical performance. According to the UK Automotive Council roadmap for power electronics [1] short term material improvements are needed to increase the temperature range, mechanical stability and robustness of conventional power packaging, as well as improved bonding mechanisms to better withstand repeated power and temperature cycling.

In the longer term, the existing methods of packaging power semiconductor devices need to be revaluated, and in particular new approaches will be needed to unlock the full potential of wide bandgap devices. Compared to silicon power device technology, their greatly increased switching speeds allow much higher switching frequencies to be achieved, leading, for example, to smaller passive component requirements, more power dense converters and a reduced bill of materials (BoM). However, such desirable features can only be achieved if circuit parasitics and associated electromagnetic interference can be reduced to unprecedentedly low levels. According to the IEEE International Technology Roadmap for Wide Band-Gap Semiconductors (ITRW) [2], this demands radically new approaches to integration, moving from assemblies of discrete components, each of which is designed and packaged separately, to fully integrated assemblies comprising power devices, gate drives, filters, sensing, and control functions.

In this paper we examine the concepts and technologies that underpin integrated power electronic conversion, including assembly methods and details of a case study CiP for an integrated drive automotive application.

The progression to Converter-in-Package (CiP) functional blocks will likely occur over differing timescales according to the power level and application. Many low-power (less than a few hundred watts) dc-dc converters are already fabricated as single-package, integrated assemblies, whereas the norm for higher power levels (greater than a few kilowatts) typically combines surface-mount components on a multi-layer PCB with larger discrete passive components and power semiconductor modules. The desire for higher switching speeds will drive a move to physically smaller commutation cells favoring increased use of smaller, surface-mount components, embedded component technologies and 3D stacked structures. Consequently, it is likely that higher power converters will be realized as several lower-power modular blocks, combined in parallel or series, so that the benefits of higher frequency switching can be maintained.

Fig. 1 shows an overview schematic for a half bridge converter building block. The key motivation is to create an optimized commutation circuit with a layout designed to minimize the commutation loop inductance. This is further facilitated with the inclusion of the DC port capacitance Ci, on the substrate. The minimized commutation loop allows the semiconductor devices to switch at their maximum speed with minimal voltage overshoot at the output port. A further consideration is the need for close-coupled, low-impedance gate drives for the power switching devices. This becomes more important at high frequencies where separations of more than a few mm can add significant parasitic inductance and result in false triggering and device destruction.

To facilitate the paralleling of these switching cells to form higher power converters, inductors are included in series with the output port, while shunt capacitors C_o provide a low-pass filtering function to reduce the Electro-Magnetic Interference (EMI) generated. When coupled with appropriate screening and shielding such an approach allows all unwanted electro-magnetic fields to be contained within the converter, reducing or even eliminating the need for external EMI control measures.

Realization of CiP solutions has generally been pursued through two distinct routes. The first of these is derived from a fusion of microelectronic packaging and laminated printed circuit board technology. It is typified by power semiconductor devices and other components embedded in a laminated build-up which is used to form interconnecting vias and tracking. The base substrate may be based on epoxy-ceramic- or insulated metal substrates with subsequent layers typically being formed using layers of epoxy-based pre-preg and copper. Interconnecting vias, for example to the die contacts, are typically formed by laser drilling and electroplating. A typical process flow is shown in Fig. 2 and a finished example of a switching cell in Fig. 3 [3].

The second approach is derived from more conventional power electronic module packaging and consists of an assembly of multiple substrates, which are stacked together to form a complete converter. Typically, the power dies are mounted onto a ceramic-based substrate (e.g. DBC, DBA), along with a limited number of passive components. Die interconnects are either wire bonded or created by planar overlays. Additional function layers, for example containing gate drives, decoupling capacitors and control circuitry are then created, typically using standard PCB assembly processes. The individual boards are then stacked together with solder pins or spring contacts making the electrical interconnections. A typical build-up is shown in Fig. 4 and its physical realization in Fig. 5 [4, 5]. In this example the power devices are SiC JFETs mounted onto a DBC substrate in a cascode configuration with Si MOSFETs. The inductor winding is mounted onto the same DBC substrate for enhanced cooling so a high current density, in excess of 50A/mm², can be utilized to keep the inductor volume to a minimum.

Finally it is clearly possible to blend the additive embedded route with the substrate assembled route, for example by embedding the power dies to create a compact functional switching cell but then adding further control functions on separate substrates. This can help to avoid some of the yield issues and reliability issues associated with complex multi-layer build-up laminates, in particular those associated with the high through-thickness coefficient of thermal expansion (CTE) of typical epoxy-based systems.

Irrespective of which die-level integration approach is adopted, one of the greatest challenges is mitigating the mismatch in CTE between the semiconductor die and the substrate and between the die and the interconnect layers.

Eutectic or near eutectic Sn-Ag and Sn-Ag-Cu solders have more commonly been used to replace the eutectic 37Pb63Sn solder for power die attachment, but they are prone to thermo-mechanical fatigue leading to early failure [6]. Au-based eutectics or near eutectics such as Au-Sn, Au-Ge and Au-Si are suitable for high reliability applications but they are very expensive and exhibit poor workability and inferior wettability.

Low temperature sintering of Ag nanoparticles is probably the most attractive technology for high temperature and high reliability application [7, 8]. Unlike solder joints, sintered silver die attachments rely on an atomic diffusion process and particles consolidation, rather than soldering based on a melting and solidification process. Therefore, sintered joints can be formed at relatively low temperature (below 300°C), but can be operated at both below and above the bonding temperature.

Historically, the sintering process has required specialist equipment and careful mechanical design to accommodate the high sintering pressures. However, with the advent of pressure-less pastes and reduced pressure and time sintering processes, it is now possible to integrate sintered die attachment into a standard pick and place process flow, for example using an enhanced die bonder [9].

Ultrasonic Al wire bonding is commonly used for manufacturing power modules but this is not compatible with a compact layered approach, it is associated with high levels of parasitic inductance and is a known reliability weak-point. Over the past years, several replacement of wire-bonds such as ribbon-bond, dimple array, embedded chip technology, silicon & ceramic interposers, solder bump, metal bump and press-pack bus-bar-like interconnects have been proposed and investigated [10–15]. These replacements have demonstrated dramatic improvements in the thermal and electromagnetic performance, and/or allow advanced integration schemes for the optimization of basic power switch topologies. Flexible printed circuit board (PCB) and CTE-constrained planar interconnects have been demonstrated not only to obtain dramatic improvement in the thermal and electromagnetic performance, but also allows for efficient and low cost manufacturing [11, 13, 16]. However, there are still concerns about the manufacturability and reliability of these techniques [14, 17].

Fig. 6 compares the experimental reliability of a range of planar interconnect technologies [17]. As might be expected, a ceramic-based interconnect with thin copper offers the highest level of reliability for solder-based joining technology, whereas those made using higher CTE flex materials yield inferior results. Using a sintered Ag to replace the solder can result in more than an order of magnitude improvement in power cycling reliability, as shown in Fig. 7.

Power electronic drives for electrical machines are typically implemented within a physically separate enclosure, although this may be mounted on the machine casing. In a truly integrated drive, the power electronics is mounted within the machine casing with advantages including elimination of machine-electronics cabling, a reduced overall volume and reduced EMI. However this comes at the expense of having to create a compact power electronics assembly that must share its thermal management with the machine and can be accommodated within the casing.

Here we present the design and assembly of a substrate assembled integrated power electronics converter for a Permanent Magnet Synchronous Machine drive consisting of three, three-phase sub-machines, an 80kW nominal power rating and a rotational speed of 30k rpm. Direct liquid cooling of the stator windings and power electronics permits a high current densities of 25A/mm² to be utilised in the machine windings. Fig. 8 shows a CAD image of the drive: the power converter consists of 9 individual modular blocks, each rated at 160A/800V and arranged around the machine periphery.

Fig. 9 shows an overview of the power electronics assembly, which is based on a stack of substrates. The top-most layer (master board) provides the master control for the three 3-phase drives. Each 3-phase sub-machine has a local controller consisting of a combination of digital signal processor (DSP) and field programmable gate array (FPGA), as shown in Fig. 10.

The remaining layers form the power conversion stages and are separated into a gate drive layer, dc-link decoupling layer, illustrated in Fig. 11 and power substrates, illustrated in Fig. 12. Each power module comprises 6-off 1200V, 98A SiC MOSFET dies, dimension 7X4 mm plus temperature and current sensors. Assembly of the 9 power substrates starts with die attach by Ag sintering followed by reflow soldering of the flex interconnect and the remaining surface mounted components, as illustrated in Fig. 13. Connections from each of the machine winding to its power substrate are made directly from the back-side and carried through vias in the substrate.

Results of initial testing of the prototype converter are shown in Fig. 14 and illustrate the exceptionally low voltage overshoot that occurs at the switching transitions. Rise and fall times of less than 20 ns were measured under hard-switched conditions.

Realizing the full potential of wide band-gap devices for automotive applications will require new approaches to their packaging and integration. The accepted separation of a power electronic converter into discrete power modules, passive components and micro-electronic controls will likely be discarded in favor of integrated assemblies based on embedded die technology and stacked substrates, ensuring the correct thermal and electrical operating environment. Results from a prototype integrated drive illustrate how a segmented, modular substrate-assembled converter-in-package approach can be used to implement a compact SiC MOSFET equipped drive for an automotive powertrain application.

The original research presented in this paper was funded by the UK Engineering and Physical Sciences Research Council (EPSRC) through research grants [EP/I038543/1] and [EP/K035304/1].