Failures of MOSFETs in Terrestrial Power Electronics Due to Single Event Burnout

Failures of MOSFETs in Terrestrial Power Electronics Due to Single Event Burnout C. Donovan Davidson P.Eng. V.P. of Research and Development Argus Technologies Ltd. 5700 Sidley St., Burnaby, B.C. V5J5E5 Canada Ewart W. Blackmore PhD Senior Research Scientist TRIUMF Laboratory for Particle and Nuclear Physics 4004 Westbrook Mall, Vancouver, B.C. V6T2A3 Canada John I. Hess Vice President, Power Products Advanced Power Technology 405 S.W. Columbia Street, Bend, OR 97702 USA Abstract - Failures of semiconductor devices caused by cosmic radiation exposure is well known in space and avionics applications. However it is not well known that similar issues exist, to a lesser extent, for terrestrial applications. We have conducted research into the phenomenon of Single Event Burnout (SEB) of high voltage power electronics semiconductor devices, specifically MOSFETs that are ubiquitously used in modern switchmode power electronics. The research revealed that SEB and related effects, caused by high-energy neutrons generated by the impact of cosmic radiation on our upper atmosphere, are a real possibility at ground level. An introduction and overview of the SEB effect, a review of existing literature indicating the flux and energy distribution of neutrons that reach the earth s surface and an overview of the structure of MOSFET devices indicating why they could be susceptible to the SEB effect is given. Experiments were conducted which demonstrated that random failures of high voltage MOSFET devices, when they are biased in the off state, do occur at a much higher rate than should otherwise be expected. A summary of experiments that were carried out at the TRIUMF cyclotron laboratory on both individual MOSFETs and on electronic equipment containing MOSFETs is presented. These experiments provide strong evidence that failures of MOSFETs and circuits using MOSFETs can be caused by neutrons with energy distributions similar to that experienced on the earth s surface. I. INTRODUCTION Single Event Burnout (SEB) is one of a family of Single Event Effects (SEE) that includes Single Event Upset (SEU) and Single Event Gate Rupture (SEGR). These effects are evident in semiconductor devices when exposed to highenergy particles such as heavy ions, protons, neutrons or pions and can result in device malfunction or destruction. The SEB effect has been recognized to be a problem for electronic systems operating in space where cosmic rays consisting of heavy ions or high-energy protons are prevalent and in avionics applications where high levels of high-energy neutrons are present. The SEB effect has generally not been recognized to be a possibility for terrestrial based electronics until very recently. Rectifiers for telecommunications powering have been using off line switchmode technology for two decades now. This technology involves the use of high voltage semiconductor devices as high frequency switches to create a high frequency AC voltage waveform that is impressed across the primary of a high frequency transformer of the rectifier. This substantially reduces the weight and size of the transformer and filters of a rectifier compared with previous line frequency designs. In addition the high frequency waveform allows the use of Pulse Width Modulation (PWM) to achieve high-speed regulation in a loss less manner to achieve very high conversion efficiency. Earlier rectifier designs such as Argus Technologies RST and RSM series used high voltage Bipolar Junction Transistors (BJT) as their main switching devices. The performance of Metal Oxide Semiconductor Field Effect Transistor (MOSFET) devices has improved tremendously in recent years and new rectifier designs use 0-7803-8458-X/04/$20.00 2004 IEEE 503

MOSFET switching devices almost exclusively. In fact BJT devices for switchmode rectifiers have become increasingly unavailable. Typically 500V MOSFETs are used in switchmode rectifiers operating with an off state voltage of 400VDC and some cases as much as 425VDC. This allows enough voltage headroom for transients, allows operation of the rectifier with good power factor at high AC line voltages and optimizes its efficiency, holdover time, power density and cost. Published data from the MOSFET manufacturers [1] indicate the MOSFETs should operate very reliably with these voltages as long as their on state current and junction temperature is kept within ratings. II. OVERVIEW OF THE SEB EFFECT Single Event Burnout is a phenomenon that can occur to a high voltage semiconductor device biased in the off state with a voltage close to its rated value. It can occur if a single heavily ionizing particle deposits sufficient energy in the depletion region of the device. This energy generates electron hole pairs and, if the local electric field is strong enough, an avalanche breakdown of the device could occur. The heavily ionizing particle can either be an energetic ion as found in space or the resultant particles generated by nuclear recoils from a proton or a neutron striking a silicon nucleus in the depletion region. Knowledge of the SEB effect has only recently been developed and has been limited to the aerospace industry, very high voltage device and nuclear science circles. The fact that heavy ions striking high voltage N-channel power MOSFETs can cause single event burnout was reported in 1986 [2]. The fact that neutrons can cause SEB of Power MOSFETs was first reported in 1996 [3] in IEEE transactions in Nuclear Science. Space and avionics power electronics have since been designed to take into account the SEB effect. The SEB effect has also been known to be a possible problem for large very high voltage devices such as used in train drives [4]. The SEB effect was introduced, as a possible cause of MOSFET failures, to the general terrestrial power electronics community by Gaylon et al in 2001 [5]. R. Sheehy et al. [6] reported in 2002 that failures of terrestrial MOSFETs displayed the characteristics of Cosmic Radiation Effects. To this point however the SEB effect has not been actually demonstrated to be a cause of failure of terrestrial power electronics equipment using MOSFETs. III. GROUND LEVEL NEUTRON FLUX Atmospheric neutrons are the main cause of single event effects in electronics at sea level and at aircraft altitudes [7]. These neutrons are created by the interaction of energetic cosmic rays (mainly protons) with the oxygen and nitrogen of the earth s atmosphere and make up over 90% of the ionizing radiation that reaches ground level. The neutron flux varies with altitude, latitude and solar activity with the flux of 1-10 MeV neutrons at sea level being about 300 times less than at 40,000 ft. Zeigler [8] gives an excellent summary of terrestrial cosmic rays and measurements of the neutron flux as a function of energy and location. For Soft Error Rate (SER) measurements the flux of >10 MeV neutrons is usually used and for New York City and similar latitudes this rate is 20 cm -2 h 1 although with a quoted accuracy about a factor of 3. Table 1 shows the neutron rate as a function of altitude based on relative measurements. Table 1. Altitude Flux Factor >10 MeV flux neutron cm -2 h 1 Sea level 1 20 5,000 ft 3-4 70 (Denver) 10,000 ft 10-15 250 (Leadville) 40,000 ft 200-300 4000-6000 Shielding that may occur within a building can also reduce the neutron flux; approximately 2 meters total of concrete will reduce the flux by an order of magnitude [8]. It should be noted that the average angle of incident neutrons is 45 degrees. IV. STRUCTURE OF MOSFET DEVICES AND THE SEB EFFECT An N-channel MOSFET device is a complex structure, as shown in cross-section in Fig. 1, incorporating N+ source, P, N- and N+ substrate doped layers. A parasitic NPN bipolar transistor is invariably present with a low value resistance between it s base emitter junction. When the MOSFET is biased in the off state with an applied voltage between it s drain and source a depletion region of approx. 10 microns per 100V, void of mobile charges, supports the applied voltage in the N- region and to a much lesser depth in the P region. This voltage is also applied to the collector to base junction of the parasitic bipolar transistor. Above 200V bipolar transistors have a Vceo rating less than their Vcer rating and for high voltage devices this rating can be significantly less. For high voltage MOSFET transistors biased in the off state near it s rating the bipolar transistor is actually operating above it s Vceo rating and would not be able to sustain this voltage should it ever conduct and could suffer second breakdown. MOSFET manufacturers endeavor to keep the bipolar base to emitter resistance as low as possible to keep the parasitic transistor off in order to ruggedize their MOSFETs. Regardless a very high electrostatic field can exist in the depletion region. The theory is that all it takes is for a single high-energy (>5MeV) neutron to hit a silicon nucleus in this region, to generate 504

sufficient mobile carriers, to cause enough avalanche current to forward bias the base emitter junction and activate the parasitic bipolar transistor [9]. It is interesting to note that P- channel high voltage MOSFETs are apparently immune to the SEB effect [9] and are known to be very rugged. Unfortunately their on state resistance is much greater than N-channel devices. Simple semiconductor devices such as very high voltage diodes can also be subject to the SEB effect [9] but to a much lesser degree while it is expected that devices with a regenerative structure such as IGBTs or SCRs will be at least as sensitive as MOSFETs under the same conditions. reliability standards. It is possible that the failures were caused by SEB of the MOSFETs induced by terrestrial neutron radiation. To prove this, and to find if failure could occur below 450V, it would be necessary to expose the MOSFETs to a known source of neutrons, with the same energy profile as terrestrial neutrons, at a much higher flux to time accelerate the experiment. Fig.2 Test Setup of biased MOSFETs in the Electronics Lab. Fig 1. Structure of Power MOSFETs. V. EXPERIMENTS WITH BIASED MOSFETs IN AN ELECTRONICS LAB. Experiments were conducted in our Electronics Laboratory to verify the possibility that MOSFETs could fail when biased in the off state at a voltage near their rated voltage. Approximately 50 samples of large die size 500V MOSFETs from many different manufacturers were placed in a test jig in our Laboratory. The test jig consisted of a 5uF capacitor placed directly across each MOSFET drain to source, with each MOSFET drain connected with a 100K resistor to a common high voltage Lab Supply. The capacitance simulates the effect a solid raw supply voltage would have on the MOSFET should avalanche breakdown commence. The MOSFET gates were connected to their sources assuring that the MOSFETs were in the off state. The experiment consisted of setting the Lab Supply to different voltages near the MOSFET rated voltage and leaving the setup at room temperature for a period of time. It was observed that some of the MOSFETs from all the manufacturers failed short from drain to source after a period ranging from 18 to 374 hrs, that no failures occurred at 450V or below and that the frequency of failures increased as the voltage approached 500V. It could be concluded from these experiments that certainly something was causing the MOSFETs to fail, when biased in the off state above 450V (90% of their rating), at room temperature. The MOSFET reliability was orders of magnitude lower than should be expected from the manufacture s data and accepted VI. EXPERIMENTS WITH BIASED MOSFETs AND OPERATING EQUIPMENT AT A NUCLEAR PARTICLE ACCELERATOR LAB It was decided to conduct experiments at the TRIUMF laboratory for Particle and Nuclear Physics located in Vancouver B.C. At the TRIUMF 500 MeV cyclotron it is possible to generate neutrons, with a similar energy distribution as terrestrial exposure, but of a flux many orders of magnitude higher. This allows a test sample to be irradiated with years worth of neutron exposure in minutes. Physicists at TRIUMF were familiar with testing for SEU using neutrons, validating the proposed experimental approach, but they had not previously looked for SEB effects. It was decided to take both the test setup of biased MOSFETs and complete operating power electronic equipment containing MOSFETs to the particle accelerator. A Rectifier was chosen for this experiment that contains 18 MOSFETs. Five test Rectifiers modified with different raw supply voltages and with MOSFETs of different voltage ratings, which had been previously operated for 12 hours to verify proper operation, were used. The MOSFET test setup and the rectifiers were consecutively installed in the proton irradiation therapy room on axis to an 115MeV proton beam line, where neutrons can be generated using a beam stop. VII. NEUTRON BEAM AT TRIUMF At the TRIUMF cyclotron a beam of energetic neutrons was produced by stopping 115 MeV protons in a lead beam stop and using the neutrons produced downstream of the beam stop [10]. These neutrons have the highest flux and energy, with a roughly uniform distribution in a cone of 505

about +/- 15 degrees on axis and away from the incoming proton beam. The resulting neutron energy spectrum is similar to that of atmospheric neutrons as the production mechanism is similar. The neutron energy spectrum had been previously measured for the test geometry using Bonner Spheres and carbon activation with the beam of 115 MeV protons (the highest energy of this particular beam line) stopping in the 20 mm lead plate. Fig. 3 shows the measured spectrum compared with a FLUKA [11] calculation. The atmospheric sea level neutron spectrum from [12] multiplied by a factor of 2x10 6 is also shown on this graph. The scaling factor has been selected to match the spectrum above 10 MeV. This means that 1 year s equivalent sea level exposure can be achieved in about 15 s at a proton current of 1 na. second section failed after a further 15 seconds of irradiation. A section of unit #2 with 500V MOSFETs operating at 385V failed after 2 minutes of irradiation (equivalent to 16 years exposure at sea level). A section of unit number 3 with one brand of 550V MOSFETs operating at 385 V failed after 16 minutes of operation (equivalent to 128 years of exposure at sea level). Units #4 with another brand of 550V MOSFETs and unit #5 with 600V MOSFETs operating at 415V did not fail despite being irradiated for 20 minutes (equivalent to 160 years exposure at sea level). The units were dismantled and it was found in all failed cases that either a PFC MOSFET or a pair of DC/DC converter MOSFETs were blown apart. In one case even a section furthest from the neutron source was found to have failed. Fig. 3 Neutron flux per na of protons as calculated by the FLUKA code and measured by Bonner spheres and activation. The corresponding sea level 6 neutron flux scaled by 2x10 is also shown. VIII. EXPERIMENTAL PROCEEDURE The test jig of biased MOSFETs containing 37 MOSFETs of 2 different manufacturers and of rated voltages between 500V and 600V was placed at right angles to the neutron beam axis 1.3 meters from the lead plate. The test jig was operated for 10 minutes at 385V, 400V and 415V consecutively with neutron irradiation at a proton current of 2nA (equivalent to 80 years exposure each time at sea level). Failures of the 500V MOSFETs occurred at all voltages with a much lower frequency of failure at the lower applied voltages, but no failures of a 550V MOSFET occurred at the lowest voltage and no failures of 600V MOSFETs occurred at any of the applied voltages. The test Rectifiers were also placed consecutively on axis with the neutron source, with their MOSFETs at right angles to the radiation, starting with a unit with 500V MOSFETs and a 415V raw supply voltage. The units were turned on and thier load was adjusted for 50A. The units were then irradiated with neutrons at the 2nA proton current. Failure of one section of the first rectifier occurred with a bang within 10 seconds of commencement of the radiation, a Fig. 4 Test Setup of biased MOSFETs in the proton irradiation room of TRIUMF. The lead beam stop and the end of the proton beam line are on the right. A Snoopy neutron detector, which gives a direct reading of the neutron flux, is on the left. Fig. 5 Test Rectifier in the proton irradiation room of TRIUMF. The proton beam line and lead beam stop is on the right. 506

conclusions can be drawn. The research into the SEB effect indicates that it is a real phenomenon that can occur at ground level. The analysis of MOSFET construction indicates that they along with similar devices such as IGBTs are particularly sensitive to the SEB effect. The failure rate that should be expected is proportional to the equipments size (area of MOSFETs) and also a very strong function of the equipment raw supply voltage (applied voltage) and the rated voltage of the MOSFETs. The failure rate will be higher for high altitude sites but lower for equipment installed in urban areas in the lower levels of multi-floored concrete buildings. It has been shown that reducing the raw supply voltage of the equipment and/or increasing the rated voltage of the MOSFETs has a dramatic effect of reducing the MOSFET failure rate. Fig. 6 Failed MOSFETs from neutron irradiation in the Test Rectifier. IX. DISCUSSION AND MEANING OF THE RESULTS The fact that individual biased MOSFETs failed while being subjected to neutron radiation indicates that the failure phenomenon is strictly associated with the MOSFETs and thier biased voltage and is not related to malfunction of other circuitry in the test equipment. The experiment in the electronics laboratory indicates that MOSFETs of all manufacturers are susceptible to this failure phenomenon. Diodes and BJT devices (even if operated above their Vceo rating with emitter open switching) will be substantially less sensitive to the SEB effect, in fact no high voltage diodes of the test rectifiers failed during the experiments. The experiments at TRIUMF provide conclusive evidence that excessive failure rate of high voltage MOSFETs and equipment containing these MOSFETs is being caused by cosmic neutron radiation even at ground level. The failure rate is high even at low device temperatures where MOSFET manufacturers reliability information indicates the failure rate should be very low [1]. It also shows that neutron radiation is very penetrating and can travel through inches of aluminum before striking a MOSFET. The experiments show that reducing the ratio of applied voltage to rated voltage of the MOSFET by as little as 6% can reduce the failure rate by an order of magnitude. X. CONCLUSIONS The combination of research into the SEB effect and experiments conducted with individual high voltage MOSFETs and equipment containing MOSFETs provides conclusive evidence that failures of MOSFETs in equipment such as telecom rectifiers is being caused by the SEB effect induced by terrestrial neutron radiation. The experiments conducted did not have sufficient sample size to be statistically accurate, so more research and experiments should be done by the research community, but regardless REFERENCES [1] S. Clement, K. Teasdale Understanding and Using Power MOSFET Reliability Data, I. R. Application Note 976A. Hexfet Power Mosfet Designers Manual Vol. 1. International Rectifier, 1993. [2] A.E. Wasiewicz et al. Burnout of Power MOS Transistors with Heavy Ions of Californium-252, IEEEE Trans. Nuc. Sci., NS-33, 1710, Dec. 1986. [3] D.L. Oberg et al. First Observations of Power MOSFET Burnout with High Energy Neutrons, IEEE Tran. Nuc. Sci., Vol. 43, Dec. 1996. [4] H. R. Zeller, Cosmic Ray Induced Breakdown in High Voltage Semiconductor Devices, Microscopic Model and Phenomenological Lifetime Prediction, Proc. 6 th International Symposium Power Semiconductor Devices and ICs, 1994. [5] G.T. Galyon et al. Static and Dynamic Testing of Power MOSFETs APEC 2001, Vol. 1, Mar. 2001. [6] R. Sheehy et al. Sea level Failures of Power MOSFETs Displaying Characteristics of Cosmic Radiation Effects. PESC 2002. [7] E. Normand, Single Event Upset at Ground Level, IEEE Trans. Nucl. Sci. NS-43, 2742, 1996. [8] J. F. Zeigler, Terrestrial cosmic rays, IBM Journal of Research and Development, Vol. 40, No. 1 pp 19-39, 1996. [9] E. Normand et al, Neutron-Induced Single Event Burnout in High Voltage Electronics, IEEE Transactions on Nuc. Sci. Dec 1997. [10] E. W. Blackmore, Operation of the TRIUMF (20-500 MeV) Proton Irradiation Facility, IEEE Radiation Effects Dose Workshop, pp 1-5, 2000. [11] A. Fasso, A. Ferrari, J. Ranft, and P.R. Sala. FLUKA: Status and Prospective for Hadronic Applications, Proceedings of the Monte Carlo 2000 Conference, 2000. [12] E. Normand, Single-Event Effects in Avionics, IEEE Trans. Nuc. Sci. NS-43, 461, 1996. 507