Solid-State Marx Modulators

Transcription

1 Solid-State Marx Modulators Dr. Marcel P.J. Gaudreau, PE, Noah Silverman, Michael Kempkes Diversified Technologies, Inc. Bedford, MA 01730, USA Dr. Jeffrey Casey Rockfield Research, Inc. Las Vegas, NV 89135, USA Abstract Diversified Technologies, Inc. (DTI) has recently completed two very different solid-state Marx Modulators one at 500 kv and microsecond scale pulses, and the second at 120 kv and millisecond scale pulses. While they are both solid-state Marx modulators, their design and construction was very different, driven by their specific requirements. As a result of our experience with these two systems, we can propose that a General Marx design is both feasible and desirable, allowing a broad range of klystron modulator applications to be met with a standardized Marx module. This paper describes each of the Marx modulators, and discusses that attributes of the General Marx design for the future. Keywords solid-state; marx; modulator, klystron I. INTRODUCTION DTI has recently designed and built two very different Solid-State Marx Modulators, described in the following sections. Although both of these systems were designed with a Marx topology, the motivations were entirely different in these two cases. The Yale Marx system, originally begun as a design for the Next Linear Collider (a non-superconducting X-band accelerator design) which requires 500 kv, 250 A, and a 1.8 s pulsewidth. This was conceived as a Marx topology system for several reasons fast risetime, avoidance of DC high voltage, and waveshaping (both leading edge waveshaping and capacitor droop correction). The short pulse allowed a common-mode choke topology for charging, thus requiring only a single switch per module (Figure 1 right). The recent long-pulse Marx system, for ILC/KEK, has a more modest specification of 120 kv and 120 A, but a very long pulse at 1.5 ms (Figure 1 left and Figure 2) for a superconducting accelerator. This was conceived as a Marx primarily for waveshaping considerations, with a minor consideration being the economy of the prime power feed. This system required far more extensive on-module energy storage, but none of the high voltage engineering of the 500 kv system. As we completed these two systems, the outline of a General Marx design, capable of meeting a broad range of klystron modulator requirements, appears feasible and economically desirable. The optimal approach is a hybrid of the two designs, providing simple, low cost construction and assembly, but capable of addressing a much wider range of applications voltages from kv, currents up to 500 A, and pulsewidths from 1 microsecond to milliseconds. Configuration of a modulator for a specific requirement would be greatly simplified, chiefly determined by selection of the number of modules and the capacitor value required per module. II. MARX GENERATOR The concept of a Marx modulator has existed for many years, and may be summarized as an array of capacitors charged in parallel at low voltage and discharged in series to form a high-voltage output. This is a legacy idea, practiced for decades using resistor charging networks and spark-gaps for discharge. Constrained by the limits of closing switches, such systems required pulse forming networks and crowbars, with their attendant limitations. One key element of DTI s design is the patented ability to construct robust series-array circuits from individual transistors effectively extending at will the voltage capabilities of today s high performance power transistors. The use of DTI s solid-state switches in place of traditional spark gaps or SCRs enables the Marx modulator to open and close nearly instantaneously, thereby allowing the capacitors to store energy between pulses, rather than fully exhausting each time the system is discharged. Figure 1. Operation of a solid-state Marx Modulator, showing the charging paths of each capactor in parallel, and erection of the Marx to high voltage by closing the switch on each stage, placing the capacitors in series. When the switch is re-opened, the pulse ends. In this short pulse example, the inductors on each stage act as a short during charging, but as an open circuit during the pulse. For longer pulses, a second switch is required to charge the capacitors in parallel.

2 Figure 2. The overall configuration of the Yale Marx Modulator, with 50 idenitcal solid-state Marx modules, oil tank, HV output connection, and controls. The speed of a solid-state array allows complete removal of stored energy from the load in less than one microsecond after detection of a fault, effectively eliminating the possibility of arc damage. As such, DTI s solid-state Marx system does not require the implementation of a crowbar, and the stored energy in the modulator can be arbitrarily large. The solid-state Marx topology also allows a new degree of freedom unavailable to other architectures we can intentionally underdamp the series snubbing within the pulse circuit. This cannot be done conventionally the reactive overshoot would endanger the klystron, causing arcs. In a Marx, we can compensate for the overshoot by initially firing only a subset of the modules thus sling-shotting the leading edge faster than otherwise possible. We can tune the number and timing of subsequent module firings to counter the reactive ringing, and hold a flattop pulse to the desired voltage and accuracy. Similarly, additional modules may be added to fire sequentially later in the pulse to compensate for capacitor droop. This is a critical enabling technology motivating Marx use for long-pulse accelerators (such as ILC), and yields valuable optimization even for very short-pulse systems. The reduction of capacitor size afforded by this flexibility further reduces parasitic capacitance, and thus reduces equipment size and cost while increasing power efficiency. III. MODULATOR REQUIREMENTS The range of potential klystron modulator requirements is very broad, extending over an order of magnitude in cathode voltage (from ~ kv). The klystron cathode voltage determines the output voltage required in the modulator. Typically, voltages higher than ~ 100 kv are addressed through the use of a pulse transformer, at the expense of efficiency and pulse shape. The other requirements include cathode current and pulsewidth, which (in combination with the voltage) determine how much energy is required in each pulse. In the chart in Figure 3, we simplify these parameters to express the range of klystron modulator requirements in terms of voltage and the total stored charge required to maintain a 1% droop during each pulse. This is a critical factor, because it determines the total capacitance required in the modulator Voltage (kv) Charge (Coulombs) Figure 3. Recent klystron modulator requirements, expressed in terms of cathode voltage (Y-axis) and the coulombs of stored energy required to meet a 1% maximum droop at the specified pulsewidth. The charge requirement is proportional to the volume of capacitors required in the modulator. (since Q = CV). This factor can vary by nearly four orders of magnitude across different klystrons, although there is a strong inverse correlation between voltage and total required charge. Designing a system capable of addressing this full range of requirements is a challenging endeavor, but appears feasible with the solid-state Marx architecture. IV. YALE MARX Under a US DOE SBIR grant and based on research begun under the Next Generation Linear Collider (NLC) program, high energy, short-pulse modulators are being re-examined for the Compact Linear Collider (CLIC) and numerous X-Band accelerator designs. At the very high voltages required for these systems, however, all of the existing designs are based on pulse transformers, which significantly limit their performance and efficiency. Prior to this development, there was not a fully TABLE 1. YALE MARX DESIGN PARAMETERS Pulse Voltage 500 kv Pulse Current 250 A Pulse Width 1.8 us Repetition Rate 20 Hz Module Voltage 12.5 kv Module Capacitance 0.6 µf # Modules for Base Pulse 40 Total # Modules in System Expected Risetime 300 ns Heater Voltage 24 VDC Heater Current 21 A Insulation Oil

3 Figure 4. The Yale Marx 500 kv modulator charges many stages in parallel at low voltage, and then discharges in series at high voltage. Each 12.5 kv, 250 A flat pack module is identical, providing for low fabrication and assembly cost. optimized, transformer-less modulator design capable of meeting the demanding requirements of very high voltage pulses at short pulsewidths. The basic design of the Yale Marx is built around 50 identical Marx modules (Figure 4), each containing the solidstate switch, capacitor, snubbing network, and local diagnostics and controls. Each module, therefore, represents an independent modulator in is own right. These modules were designed in a flatpack configuration, allowing simple manufacturing and assembly. A. Reduced Module Costs Through the use of PC board trace shielding and RF cans in sensitive areas of the circuit, we were able to co-locate controls directly on the board within reasonable proximity to pulsed current sections of the same board. By exposing the IGBTs directly to the oil, we can cool the devices effectively while eliminating machined parts and hardware, further reducing the module parts count, associated materials, and assembly costs. Since the flat-pack design significantly reduces voltage gradients from module to module within the Marx bank, the only significant need for corona and field reduction geometry is at the interface between the Marx bank stack and the walls of the tank. These factors enable a significant reduction in module costs, with the potential for additional reductions in manufacturing costs, compared to earlier designs. B. Simpler Interconnect Each module in the redesigned Marx bank plugs directly into the two adjacent modules. We oriented the connectors and offset the board components and corona shields to allow any individual module to be added or removed from the stack, much like a book on a shelf, without any significant disassembly of support structures or disruption of the rest of the Marx bank. Since all module electrical connections and buswork are integral to the modules, only the fiber-optic control line (and optional fiber-optic monitor) needs to be externally connected. The first module in the stack plugs into a connector supplying charging HV, core bias current, auxiliary housekeeping power, and ground. A connector on the final module is connected to the Marx output coaxial cable, and includes a loopback for the common-mode choke bias current. C. Scalability The flat-pack Marx design is inherently modular and scalable, as additional plates may be added for a wide range of voltages. Whether at 100 or 500 kv, a Marx bank can use the same modules. The primary impact of additional plates is an increase in the charging current at the first plate, since it carries the current for all subsequent plates as well. Stray capacitance is also greatest at higher voltages, assuming constant spacing to the tank. Figure 5. A test showing four Yale Marx modules commanded sequentially to fire. As each module is turned on, the output voltage increases by the module voltage. This independent triggering capability is unique to solid-state Marx Modulators at this time-scale, and allows precise pulse shaping and control. Figure kv Yale Marx Pulse. Initially, 30 modules were all turned on simultaneously, and the impedance mismatch with the load resulted in significant overshoot and ringing. When 9 of the modules were delayed in their turn-on, a nearly ideal pulse was obtained, The ability to control each module s timing independently provides significant flexibility in pulse shaping.

4 V. PROTOTYPE TESTING During testing of the prototype Yale Marx modulator, we ran four modules with all four modules simultaneously switched for a 1.5 µs pulse, with a 10 kv pre-charge into the system, resulting in a 40 kv output (Figure 5). The prototype modules performed flawlessly into high voltage, with none of the chatter or jitter that would otherwise be associated with gate drives impacted by noise coupling at high voltage. Most recently, DTI has tested a stack of 29 plates into a resistive load (Figure 6). The result was a 2 µs pulse output of approximately 280 kv and 185 A. Construction and test of all 50 modules is now complete, with final assembly and test planned in Fall VI. ILC/KEK MARX MODULATOR. Originally conceived, designed, and built under a US DOE SBIR grant to support SLAC, DTI has implemented a solidstate Marx topology to enable a new approach to compact and inexpensive ILC-class performance (long-pulse, high voltage). The design meets or exceeds all ILC specifications in a smaller form-factor than other known technologies (Figure 7). There are a number of approaches available for the charging of the capacitors at low voltage. Resistive isolation of the capacitors often proves sufficient in the case of very low duty cycle, as losses are limited. In the case of short pulses, inductive isolation of the capacitors proves attractive. Neither approach is suitable for the long pulses required of the ILC; instead, each capacitor is separated by two switches one for charging, the other one for pulsing. It is important to note that the gross quantity of switching silicon is relatively constant for applications such as ILC every solid-state modulator requires the same peak power switching capability, regardless of its architecture. The advantages of the Marx topology lie not in reducing the cost of the HV switches, but in allowing flexibility of the system design in order to optimize secondary characteristics. We find that this flexibility is particularly powerful for the ILC-class specifications substantially reducing the volume and cost of a high-reliability transmitter. A. Topology The ILC/KEK Marx modulator is comprised of four primary components: The Input Buck Regulator. Unregulated DC power, typically from a transformer/rectifier, is regulated to produce a prime power of approximately 6 7 kv DC. Core Switch Modules. Twenty highly efficient core switch modules erect input power to produce a pulse with voltage twenty times that of the prime DC feed. Low Power Buck Regulator. This buck regulator provides power to the sixteen corrector modules, stepping down the input from 6 7 kv DC to 900 V max for trimming. Corrector Switch Modules. Sixteen corrector switch modules take 900 V and are discharged at staggered intervals to account for voltage drop as the capacitor bank droops. This approach optimizes efficiency and maintains low cost, as the staggered discharge of the corrector module removes the requirement of a large capacitor bank for limitation of droop. B. System Integration The Marx Modulator is installed in an oil tank with a 53 x 88 footprint, and 52 in height (Figure 8). Controls reside in an oil free doghouse on the tank top, which holds two 6U standard rack drawers in height, by two racks in width. The interior of the tank is quite dense; the core modules comprise one entire face of the modulator interior, and the correctors and both buck regulators the other face. At the far end of the tank, an output snubbing circuit is included, which is no different than similar snubs used in hard switch modulators. This consists of a small inductor to limit di/dt in the case of a load arc, along with a freewheeling diode Figure 7. DTI s ILC-Class Marx Modulator in tank, installed and tested at KEK in Japan in October *Work supported by U.S. Department of Energy SBIR Award DE- SC Figure 8. The ILC/KEK Marx system, showing the 20 core modules which comprise half the tank volume. Each module used fourteen series 500 V 4700 µf electrolytic capacitors for effective module storage of 335 µf at 7 kv. Over 80% of the volume of the core modules is used by these capacitors.

5 and damping resistor across the inductor. Diagnostics of output voltage and current are also included in the tank on the output section. C. Installation at KEK Following shipment from SLAC, the ILC-class Marx modulator was successfully installed at KEK in late October, Initial tests were run at approximately 120 kv and 80 A over 1.5 ms. A representative pulse is shown in Figure 9. VII. ILC/KEK SUMMARY The solid-state-enabled Marx designed and built by DTI provides an inexpensive and compact ILC Modulator design. The precision control inherent to solid-state switching removes the need for large capacitor banks to correct for droop, instead implementing individual corrector modules to supplement the primary output pulse as voltage falls. Similar precision in the core switches eliminates the possibility of arc damage to the circuit, removing energy from the load in less than a microsecond after detection. Finally, the pulse control granted by DTI s solid-state Marx design allows system operators the unprecedented ability to tailor the pulse rise time and flattop for optimal RF performance. The ILC Marx Modulator has performed as designed. It promises to be a reliable, high performance, and economical solution to the needs of long-pulse modulators. VIII. GENERAL MARX DESIGN As a result of the lessons learned in constructionof these two application-specific solid state Marx Modulators, it is now possible to envision a generalized Marx design. We start with the simple flat pack construction of the Yale Marx, and make three major modifications replacing the common-mode choke with a charging switch, similar to the ILC/KEK Marx design; moving the capacitors from within the Marx module to between the modules (Figure 10); and implementing a selfpowering scheme for the controls, to eliminate the magnetics present in both the Yale and ILC/KEK designs. This new module design allows us to plug-in capacitors of various size as needed for systems of different pulse-length requirements, Figure 9. Pulse from the ILC Marx following installation at KEK. Pulse is 1.5 ms, 114 kv, 74 A. Ch1 (Blue): Voltage, 15 kv/div. Ch2 (Cyan): Current, 8 A/div. Ch3 (Magenta): Analog address, each step represents an additional corrector being added. When tuned, the steps are evenly spaced. Note the start is a 20x step. Ch4 (Green): Command. Figure 10. The General Marx will leverage the Yale Marx flat pack design, but the capacitor will be separated from the board and sized appropriately to varying system needs. The system in this figure has 14x the capacitive energy of the Yale Marx system, with less than twice the total volume. without modifying the Marx module itself. It is straightforward to specify different capacitors with their length and width fixed, requiring only their depth to vary for different requirements. For Marx systems, we charge up the stack, thus truly fitting the description of charging all capacitors in parallel, discharging them in series. For short-pulse systems, such as the Yale Marx, we can use a common mode choke to apply low impedance charging currents between pulses (differentially within the choke), while applying high inductance common mode impedance across the choke during the pulse. This is an elegant solution, allowing both control power and charging current to be distributed to all of the modules, along with heater power for the klystron. Unfortunately, it requires a very high quality choke design for suitable performance, and this choke must be redesigned for each new configuration of voltage, pulsewidth, or current, significantly changing the module layout for each new requirement. Clearly, this is unacceptable for a widely applicable design.. The alternative charging scheme requires a second active switch on each unit cell, as implemented in the ILC/KEK design. This second switch provides the low impedance pathway for charging currents between pulses, yet blocks full voltage erected during the pulse. Although this may not be optimal for the ultra-short pulse systems, it will eliminate much of the new engineering necessary for each set of system parameters, and will enable a common Marx module with much broader applicability, at the expense of a second, full voltage switch. Optimizing the cost of this charging switch moves us to higher voltage Marx modules than either the Yale or ILC/KEK designs, which are around 6 and 12 kv respectively. The reason is that the DC current rating of typical IGBTs will require larger, more expensive devices for modulators at high

6 average power (e.g., kw), where the charging switch must carry up to 60 A of DC current in the first (ground level) stage. It appears that (up to) 30 kv stages will be more economical in the general Marx design, since a 300 kw average power system will require only 10A of charging current, enabling smaller and cheaper devices to be used in the charging switch. Highe voltages might be possible, but 30 kv will require a relatively large module, and is also approaching the largest DC voltage that could be built for reliable operation in air, without substantial attention to corona in its construction. The third major modification is the method to transfer control power for each stage. This power is required for the gate drives, diagnostics, and other non-power circuitry, and is typically at under 15 V and just watts. This power must be present in each module even as the module is erected to high voltage during pulsing of the Marx modulator. In the Yale Marx, we use the charging inductor to pass control power to each stage, but we are eliminating this inductor in the General Marx design. Similarly, in the ILC/KEK design, a dedicated choke is used to pass control power to each stage (which is not in the pulse path), but this is impractical at at the highest voltages which we want to address ( kv). Therefore, we plan to derive control power directly from the capacitor on each stage. Finally, for long pulses (Q is > ~10 in Figure 3), capacitor storage alone is infeasible for size and cost reasons. In these cases, we will use a solid-state regulator as the first stage in the Marx. This regulator compensates for capacitor droop of up to approximately one full stage voltage. The operation of this novel regulator design developed by DTI is shown in Figure 11. In effect, the regulator acts as an infinite capacitor within its voltage authority, reducing the actual capacitor size required in each Marx module significantly. For example, if the requirement is for 1% droop, and the regulator can provide just 10% authority (e.g., 10 kv for a 100 kv system), the capacitor value and volume can be reduced by an order of magnitude over a system without the regulator. This can keep the overall modulator volume reasonable, even at very high voltages and long pulses. A key advantage is that, unlike several alternative long-pulse modulator designs, the infinite capacitor requires no additional energy, and only a fraction of the total pulse power must be switched at high frequency (just 10% in the example above). The resulting General Marx design, therefore, can be based on several key elements: A standardized Marx module at kv, applicable to the majority of klystrons simply by varying the number of modules; Selection of the appropriate Marx module capacitor size based on the current and pulsewidth requirements, without the need to modify the other design elements; Addition of an active regulator stage at the ground end of the Marx modules (where appropriate) to reduce the overall size and/or cost of long-pulse systems; Design of the Marx modules to operate in either oil or air to meet the preferences of each user facility; and Use of standard, commercial power supply designs at the Marx module voltage to charge the Marx and provide the average power required by the klystron. The combination of these elements, we believe, will enable a very affordable, easily configurable modulator suitable to a wide range of future klystron modulator requirements. V (kv) t (ms) Figure 11. Operation of a switching regulator during long-pulse operation. Initially, the high frequency regulator is charged in opposition to the main capacitor, then transfers its energy in support of the main capacitor during the pulse. By placing the zero-voltage point in the center of the pulse, no net energy is required by the regulator making it act as an infinite capacitor. In a Marx implementation, the regulator would operate as the lowest stage of the Marx (at ground level).