FPGAs use AI for soft switching

Hard switching is the most commonly used technique for DC/AC power converters but it has numerous known drawbacks. Switching losses represent the wasted energy produced during the transistor’s transition between on and off states. Such losses can account for over half of the inverter losses. In contrast, soft-switching minimises switching losses but is has never been successfully implemented for DC/AC systems. This is due to varying duty cycles, input voltage, temperature, and load conditions. Now, Silicon Valley start-up, Pre-Switch is employing artificial intelligence (AI) techniques to constantly adjust the relative timing of elements to ensure resonant transitions over all normal operating conditions, thereby realising the promise of commercial grade soft switching for the first time.

Pre-Switch re-examined the challenges presented by zero voltage switching (ZVS) and zero current switching (ZCS) through the lens of AI from the field of robotics. This new approach made it clear that ZVS is a classic intelligence application for two main reasons. Firstly, the system has limited access to parameters in a noisy environment and therefore must operate with degrees of uncertainty. Secondly, as reactive behaviour can never be ‘on time,’ the system must pre-act. Signals must launch without a well-defined stimulus to indicate when to act. Making ZVS a reality, therefore, requires a system that is statistical in adaptation and predictive in nature; an ideal application for AI.

FPGA with embedded AI

The company has developed the Pre-Flex FPGA with embedded AI, which controls the timing of auxiliary resonant switches, used in conjunction with inductors, and capacitors to ensure that ZVS or ZCS soft-switching continues independently of changes in bus voltages, temperature changes, device tolerances, device degradation or load currents.

 Pre-Flex also incorporates Pre-Switch Blink, a set of advanced safety features that monitor device performance and continuously communicate the data through an integrated serial port. Pre-Flex learns, remembers, and adjusts in-system on a cycle-by-cycle basis, ensuring clean and accurate soft-switching despite changing parameters such as input voltage, load, device tolerances, device degradation, system, and device temperatures. The result is the introduction of true soft-switching for DC/AC applications as well as substantial improvements in AC/DC soft-switching. Switching losses are reduced by around 80% in IGBTs and up to 95% in SiC and GaN MOSFETs enabling five to 20 times faster switching frequencies.

The architecture also enables reduced dV/dt (dU/dt), reduced common mode noise (that is known to cause motor bearing damage), and a reduction in the resultant EMI as well.  This effective elimination of switching losses combined with reduced dV/dt opens a new frontier for system-level improvements in cost, efficiency, power density, motor efficiencies, and motor reliability.

There are multiple dimensions to Pre-Switch’s AI that are needed to enable zero voltage switching or zero current switching in applications where changing duty cycles, loads, input voltage, system and device temperature, changing device tolerance and device degradations are all compensated for to enable a successfully optimised forced resonant switching event. Data captured from each switching cycle is stored and the results are compared to the expected result. If the results are not optimal, specific timing adjustments are made to ensure the next cycle will be fully optimised.  If the results are significantly outside of the predicted results, the error is flagged, and the system is intelligently shut down. The result is consistently reliable soft-switching, even with wide input, system, and load variability.  The FPGA also intelligently uses its understanding of the system parameters to reduce the amount of dead time needed in the system and increases system safety.

Benefits to power converters

Pre-Switch technology dramatically reduces switching losses in IGBT’s and SiC MOSFETs. This reduction of switching losses enables two new degrees of design freedom for design engineers:

When switching frequency (Fsw) is kept unchanged, Pre-Flex technology increases the efficiency of the inverter.  When the technology is used to increase the Fsw to maintain existing inverter efficiency, passive components used in filtering the output sine wave can be reduced in size, weight and cost significantly.  In the case of a motor inverter, the output current ripple can be dramatically reduced, which decreases associated motor iron/magnetic losses. The increased ratio between Fsw and the electrical fundamental frequency can decrease the total harmonic distortion (THD) of the phase current waveforms. The result is superior battery-to-driveshaft system efficiency in electric vehicles (EV’s). This increases battery range and motor torque for the same input power, all with a cooler, more reliable motor.

Pre-Switch technology enables each of these options or any combination of them. The company also enables the advantages of a low dV/dt (dU/dt) in the motor despite potential increases in switching frequencies, which benefits motor insulation, motor bearing reliability, and reduces EMI. Preliminary research suggests that Pre-Flex technology can extend EV range by five to 12% and significantly improve the efficiency and lifetime of industrial motors.

Soft switching technology

Figure 1 is a schematic of the forced resonant soft-switching circuit topology known as auxillary resonant commutated pole (ARCP) using MOSFETs. Note, the external anti-parallel diodes are shown for clarity.

Figure 1: ARCP half single-phase schematic

ARCP was invented by General Electric in the late 1980s. It promised to reduce switching losses significantly by either zero-voltage or zero-current switching, ZVS or ZVC, respectively. The fundamental obstacle to its adoption was an inability to control the topology with varying input, output, temperature, and device behavior conditions. It was not until the introduction of Pre-Flex technology that forced resonance soft switching has become commercially viable.

An ARCP circuit is used for each phase leg in power converter designs, as shown in Figure 1. The circuit consists of resonant devices C1 and C2 (the sum of which is Cr), Lr and active switches A1, A2, and diodes DA1 and DA2 shown above. It is important to point out that the additional resonant devices are small and low in cost when compared to the main switches S1 and S2.

As shown in Figure 1, Pre-Switch uses low cost IGBTs for resonant switches on both IGBT and SiC/GaN MOSFET-based systems. The resonant switches are generally selected to conduct around 1.5 times the peak load current, but the switches only need to carry this current for a minimal duty cycle (one to 5%) and hence only need to be pulse rated - which further reduces the cost of the resonant switches. Capacitors C1 and C2 are typically cost-effective ceramic capacitors between two to 200 times relative to the power device output capacitance at rated voltage. The resonant inductor Lr is also surprisingly small and low cost. Lr varies with parameters such as voltage, Fsw and resonant current, among others.

Conclusion

Pre-Flex AI-based technology meets all current definitions of AI and offers a quantum leap forward in switching loss elimination and fast acting system level safety. The algorithm monitors changing system environmental parameters and uses its adaptive in-system learning to intelligently converge, on a cycle-by-cycle basis, on the precise timing necessary to ensure reliable and accurate soft switching.

The switching loss savings can be apportioned between improved and lower cost converter efficiency or faster switching frequencies for improved load efficiency or cost, depending on the desired system result.  It is the significantly faster switching frequencies that reduces the cost of output filters and offers larger battery-to-wheels efficiencies in EVs.

PCIM Europe Visit Pre-Switch – Hall 6-119