AI-based Soft-switching Enables Efficient Power Conversion

Article By : Maurizio Di Paolo Emilio

Pre-Switch delivers efficiency and performance benefits to a wide range of applications, including electric vehicles and renewables.

The state of the current power electronics is increasingly focused on reliable power converters that can reduce the costs of the entire system. Over time, the engineering approach has been directed towards leakage removal, using higher performance semiconductor materials than silicon, such as SiC and GaN.

New materials, innovative packaging, and advanced digital control techniques help engineers and device manufacturers to improve conversion efficiency, reduce power loss, weight and costs.

According to the latest Global Solar Demand Monitor published by GTM Research, annual solar system installations will remain above 100 GW by 2022. It is clear that the growth in electricity production from solar PV should be further boosted to meet the demands of an increasingly clean planet. In any case, all that power must be processed, controlled and distributed, and reconverted by power electronics and power semiconductors.

In addition, the drastic cost reduction of lithium-ion batteries has opened up a vast new market for power electronics, represented by the automotive electromobility revolution. Analysts estimate a significant installation of batteries with costs increasingly reduced.  A recent study published by McKinsey predicts that the annual demand for lithium-ion batteries will reach 2900 GWh by 2030 (figure 1).

Figure 1: Annual Li-ion battery demand in GWh and cost trend in USD (Source: Pre-Switch)

Hard and soft switching

When the transistor is on or off, the transition time needed to reach the next working state is very short, but it is not instantaneous and produces a waste of energy losses (switching losses). Switching losses are responsible for a large percentage of the power converter losses.

Hard-Switching is simply forcing the transistor to turn on and off by adding current or voltage to enable the modified states. Hard switching is known to be very hardware demanding on transistors and shortens their life span.

Power converters using Hard-Switching must balance the increase in switching frequencies with the need for losses to meet desired system efficiencies. In practice this means that systems that require high efficiencies have to switch slowly to gain efficiency.  Designers must employ larger energy storage solutions to maintain power for a longer period of time between transistor switching cycles.

The reduction of the switching frequency implies an increase in harmonic distortion, resulting in the use of output filters.

In practice, hard-switching limits the maximum working switching frequency of transistors. Transistors have a maximum operability in terms of heat to dissipate which must be managed effectively between the various losses involved. Increasing the switching frequencies to reduce the size of a system means that the transistor must carry less working current to withstand the higher switching losses. This can be solved by adding a larger transistor with additional costs to the system. Without the switching losses the transistors would be free to switch much faster or handle more current for high-power applications (figure 2).

The concept of Soft-Switching, on the other hand, is to use an external circuit to avoid the overlapping of voltage and current waveforms when switching transistors. There are two types: self-resonant and forced resonance. In the first case there is a self-oscillating circuit and this results in a reduction in switching losses, an increase in efficiency and a reduction in electromagnetic interference. The application disadvantage limits it in the power converter market for DC/DC converters.

Figure 2: hard-switching architecture (Source: Pre-Switch)

Forced resonance soft switching topologies have the same advantages as the previous one, but are computationally demanding, cumbersome and with limited adaptability to different input conditions and load ranges.

AI for the switching technique

In recent years many AC/DC, DC/DC, DC/AC solutions have focused on the development of faster switching devices with lower conduction losses and the development of new switching topologies. IGBTs are still a standard used in various converter solutions, with SiC and GaN becoming more and more prevelant  as costs are reduced.  There are many available layout technologies, and engineers can optimize their solutions according to the application.

Field Stop Trench IGBTs offer a significant improvement in terms of loss reduction. Most of the latest generation IGBTs from leading manufacturers use combinations of structure geometry to allow for optimized energy concentration.

However, material limitations and additional implementation costs for newer and more sophisticated manufacturing processes still represent a challenging barrier to optimal system efficiency improvement with traditional components.

In high-voltage applications, the use of GaN and SiC solutions are growing in popularity, because they offer reduced switching losses and therfore  the option of increased switching frequencies. The immediate impact of the increase in operating frequency would have a tangible effect on the solar inverter market, for example, with what could be a drastic reduction in output inductor size, weight, and cost.

Increasing the frequency implies the need to contain noise and its transients. Large-scale use of new power switches could remain out of reach if the operation of power converters remains tied to traditional switching architectures.

“By reducing the frequency, we enter the soft-switching market. Soft-switching is still only used in self-resonating DC/DC power converters. Isolated soft-switching DC/AC power converters have never been perfected, which is why energy engineers call soft switching for high power AC/DC the “Holy Grail” of power electronics,” said Bruce Renouard, CEO at Pre-Switch Inc. However, simply increasing the transistor transition times with faster devices results in intolerable levels of dV/dt and EMI.

Pre-Switch has solved the problems of soft-switching by employing a built-in Artificial Intelligence (AI) integrated circuit (called Pre-Flex) that precisely controls and adjusts the timing of a very small, low-cost resonant circuit to ensure minimal overlapping of the current and voltage waveforms of the switching devices.

Soft-switching with built-in AI enables a 70-95% reduction in switching losses and solves dV/dt problems associated with faster transistors.

“Pre-Switch guarantees accurate soft-switching and reduced EMI, at higher switching frequencies than ever before,” said Bruce Renouard

The Pre-Flex integrated circuit learns and adapts to the changing system inputs and device conditions on a cycle by cycle basis to ensure optimal soft switching. In practice, it locks each transistor into reliable forced resonant soft-switching despite variations in input voltages, output loads, system temperatures, and manufacturing tolerances (Fig. 3).

Figure 3: Pre-Switch architecture (Source: Pre-Switch)

The technology has been used to switch 600V IGBT transistors at more than 100kHz and 900V silicon carbide transistors at 1 MHz. The addition of this device has insignificant costs savings when compared at the system-level.  Additionally, Pre-Switch technology can be used to upgrade existing hard-switch systems in the field. Pre-Flex has been integrated into a standard driver board for a 1200V 225A EconoDUAL in a half-bridge configuration.

“Pre-Flex is designed to work with either a half-bridge, full-bridge or three-phase configuration power converters. Each IC includes a built-in serial communications port to communicate fault conditions and also includes Pre-Switch Blink™, which ensure maximum safety features on a cycle-by-cycle basis. The Pre-Flex IGBT family is frequency-limited to 100 kHz and typically eliminates 70-85% of system switching losses. The Pre-Flex SiC/GaN family is frequency-limited to 1Mhz and typically eliminates 90-95% of total switching losses in the system including the overhead of the extra devices.  Additionally, the architecture and has a built in lossless dV/dt filter,” said Bruce Renouard.

Conclusions

The use of Pre-Flex has shown a clear improvement in the main parameters, as shown in Table 1. X-Factor is a normalized coefficient that Indicates how many times faster a device can be switched using Pre-Switch AI control algorithm  technology for the same losses when compared to the same device being hard-switched. This factor provides an indication of improved performance in terms of both current and switching frequency.

Table 1: Data analysis with consequent improvements in the Pre-switching technique (Source: Pre-Switch)

“Pre-Switch is enabling customers to build systems with switching frequencies 4X-5X faster than their hard-switched IGBT systems and 35X faster than their hard-switched SiC and GaN systems: this is achieved with half the transistor count. In the case of a SiC-based EV inverter, increasing the switching frequency from the ubiquitous 10kHz up to 100kHz or 300kHz creates a near-perfect sine wave without any output filter.  The result is the elimination of unnecessary motor iron losses and increased motor efficiency at low torque and low RPM.  Higher switching frequencies also enable higher RPM motors that are lighter and lower cost,” said Bruce Renouard.

Figure 4: Signal Analysis and AI Control Behavior (Source: Pre-Switch)

The CleanWave 200kW silicon carbide (SiC) automotive inverter evaluation system enables power design engineers to investigate the accuracy of the company’s soft switching architecture and platform over varying load, temperature, device tolerance, and degradation conditions. The platform includes the Pre-Drive3 controller board, powered by the Pre-Flex FPGA, and the RPG gate driver board, together which virtually eliminates switching losses, enabling fast switching at 100kHz.  Double pulse test data demonstrated that the Pre-Switch soft-switching platform reduces total system switching losses by 90% or more (figure 4).

At the first switching cycle 0 (corresponding when the “T” in the preview screen at the top left of figure 4), the AI Pre-Switch controller evaluates multiple inputs and decides which mode the system is in and then makes a safe but not optimized estimate of the resonance period needed for soft switching. All inputs and outputs are accurately measured and stored for future learning. The AI will finely optimize the entire system after the completion of another teach cycle.

In switching cycle 1, all AI inputs and outputs resulting from switching cycle 0 are again accurately measured and analyzed.  The IA will again output the second period of conservative resonance time similar to switching cycle 0 to ensure safe but not optimized soft-switching.

Subsequently, the AI algorithm predicts the optimized resonance time to ensure complete soft-switching with minimal loss in all aspects of the system. In subsequent stages, the system compares system inputs and the results of previous switching cycles and adjusts the resonance time to fully optimize soft-switching with the increasing load current (blue line).

System temperature changes, device degradation, and sharp current fluctuations are all considered and optimized within the Pre-Switch AI algorithm.

“Compared to traditional topologies, Pre-Flex has demonstrated a drastic reduction in switching losses (70-95%), a reduction in EMI, and a reduction in dV/dt. The technology allows low-cost IGBTs to compete favorably with more expensive technologies such as SiC MOSFETs and allows SiC technologies to switch up to 20 times faster than they do today, all while solving dV/dt and EMI problems generated as a by-product of hard-switching architectures,” said Bruce Renouard.

The topology and Pre-Switch control algorithm can provide broad-spectrum performance, offering an overall envelope for power loss reduction depending on the different operating points in each type of application.

Leave a comment