Nuclear Instruments and Methods in Physics Research B

Nuclear Instruments and Methods in Physics Research B 268 (2010) 2092 2098 Contents lists available at ScienceDirect Nuclear Instruments and Methods in Physics Research B journal homepage: www.elsevier.com/locate/nimb Nuclear microbeam studies of silicon germanium heterojunction bipolar transistors (HBTs) q G. Vizkelethy a, *, S.D. Phillips b, L. Najafizadeh b, J.D. Cressler b a Sandia National Laboratories, Albuquerque, NM, USA b Georgia Institute of Technology, Atlanta, GA, USA article info abstract Article history: Available online 26 February 2010 Keywords: IBIC SEE SiGe HBT Nuclear microprobe SiGe HBTs are very attractive devices to be used in space communication applications. This technology combines the high speed of the III V semiconductors with the well-established and easy manufacturing processes of silicon, which allows the manufacturing of RF, analog, and digital devices on the same wafer. In addition, SiGe HBTs were found to be extremely radiation hard in the context of total ionizing dose and displacement damage. However, it was shown through experiments and simulations that these devices are vulnerable to single event effects (SEEs). SEEs are changes in the normal operation of the device (its logical state, currents, transients, etc.) due to the induced currents in the electrodes by the movement of carriers created by the incident ions. We used four electrode (base, emitter, collector, and substrate) IBIC measurements at the Sandia Heavy Ion Nuclear Microprobe Facility. SiGe HBTs are usually designed using deep trench isolation (DTI) to minimize parasitic capacitances from the subcollector to the substrate (faster speed), as well as allow devices to be fabricated much closer together. It is an added bonus that the DTI does not let carriers from outside hits to diffuse into the junction and induce current. Our experiments and TCAD simulations showed that while the above goal was accomplished by this design, it increased the amount of induced charge for ion hits in the active area. Single event transients (SETs) were investigated in both standard and radiation hardened by design (RHBD) bandgap voltage reference (BGR) circuits. Ó 2010 Elsevier B.V. All rights reserved. 1. Introduction Silicon has been the favorite material of the semiconductor industry for many decades. Large defect-free crystals can be grown easily with very high purity. It is easy to dope either p-type or n-type to make various devices on the same wafer. The most important advantage of silicon is that a very high quality dielectric, silicon dioxide, can be simply grown with a variety of methods. This allows manufacturing of analog (bipolar junction transistors, BJTs) and digital devices (field effect transistors, FETs) on the same wafer, leading to high level circuit integration and miniaturization. Although silicon seems to be the perfect material for integrated circuits, it has its limitations. Its carrier mobility is relatively low and being an indirect bandgap semiconductor, it has low light emission efficiency. Today s electronic industry requires high speed and high level of device integration at low cost. Higher speed can be q Sandia is a multi-program laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy s National Nuclear Security Administration under Contract DE-AC04-94AL85000. * Corresponding author. E-mail address: gvizkel@sandia.gov (G. Vizkelethy). achieved by using compound semiconductors such as GaAs and excellent optical properties can be achieved using bandgap engineering by adding In, Al, or other elements to GaAs. Unfortunately, these semiconductors do not have a robust, easily grown oxide. In addition, the compound semiconductor industry has significantly lower yields than the silicon industry; therefore, higher cost is associated with these devices. The space industry requires that electronic parts used in satellites and other spacecraft be radiation hard. Radiation hardened devices usually have lower speed, higher cost, and larger size than commercial off-the-shelf (COTS) devices. A new semiconductor technology, which has higher speed than traditional silicon devices but uses the well-established silicon manufacturing method, is silicon germanium heterojunction bipolar transistors (HBTs). The addition of Ge to the base of a standard Si BJT allows for bandgap engineering and using the Si manufacturing processes, analog HBTs can be combined with digital CMOS devices on the same wafer. An excellent review of the SiGe technology can be found in [1]. SiGe HBTs can compete with compound semiconductors and they were found to be radiation hard to total dose and displacement damage. However, they seem to be sensitive to single event effects (SEEs). SEEs are caused by the induced currents on the device terminals 0168-583X/$ - see front matter Ó 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.nimb.2010.02.016

G. Vizkelethy et al. / Nuclear Instruments and Methods in Physics Research B 268 (2010) 2092 2098 2093 Fig. 1. Typical cross-section of a SiGe HBT after [1]. due to a single ion hit. When ions travel through a semiconductor, they create electron hole pairs by ionization. These carriers move by drift and diffusion processes in the device and can induce current in the electrodes. This current then can propagate through circuits resulting in various SEEs. The most frequent SEEs are single event upsets (SEU) when a bit of a memory array flips due to the ion hit and single event transients (SET) when a current pulse propagates through the circuit resulting in undesirable behavior. More serious SEEs can occur such as single gate rupture or single event burnout, which render the device unusable. Experiments [2,3] and device and circuit simulations [4] showed that SiGe HBTs are prone to SEEs. An excellent review of radiation effects in electronic devices and SEEs can be found in [5]. To understand the mechanism of SEEs it is necessary to study the ion beam induced charge/current (IBIC) [6] in devices and circuits. In addition to helping to understand the SEE mechanism, IBIC and Time Resolved Fig. 2. Induced charge maps for the DT (a) and non-dt (b) devices. Fig. 3. IBIC spectra (from the collector) of the DT and non-dt devices.

2094 G. Vizkelethy et al. / Nuclear Instruments and Methods in Physics Research B 268 (2010) 2092 2098 Fig. 4. Simulated induced current (a) and charge (b) in DT and non-dt devices. IBIC (TRIBIC) can help validate and calibrate device simulation (TCAD) codes. A review on the use of heavy ion microprobes in failure analysis of microelectronic devices can be found in [7]. We used the Sandia National Laboratories (SNL) nuclear microprobe facilities to perform IBIC and SET experiments on various SiGe HBT devices and circuits. We investigated how the deep trench isolation (DTI) affects the charge induction properties of SiGe HBTs and studied the SET response of SiGe HBTs and a SiGe bandgap voltage reference (BGR) circuit. 2. Experimental The SiGe HBTs used in these experiments basically have the same structure as shown in Fig. 1. In the IBIC experiments we used two devices, one with a DTI of 8 lm (DT device in the followings) and one without DTI (non-dt device). 36 MeV oxygen beam was used in all the experiments focused into a 0.8 lm by 0.5 lm spot. The beam was electrostatically scanned over areas ranging from 15 15 lm 2 to 100 100 lm 2. In the IBIC measurements the HBT s all four electrodes (base, emitter, collector, and substrate) were connected to Ortec 142A charge sensitive amplifiers and Ortec 671 spectroscopy amplifiers. These signals were connected to a FastCOM MPAWIN multiparameter data acquisition system where the four IBIC signals were recorded in coincidence with the x and y positions of the ion beam. The data was recorded in list mode and processed off-line later. All the electrodes of the HBT were grounded except the substrate, which was held at 4 V modeling the worst-case bias scenario. The induced charge was calibrated

G. Vizkelethy et al. / Nuclear Instruments and Methods in Physics Research B 268 (2010) 2092 2098 2095 Fig. 5. Circuit diagram of the tested BGR circuit after [12]. Fig. 7. Top view and cross-section of the tested standard (a) and RHBD (b) devices after [12]. Fig. 6. The physical layout of the tested BGR circuit after [12]. by using a Hamatsu PIN diode, which has about 300 lm of intrinsic layer, so all the charge created by the oxygen ions was deposited in the sensitive layer. The SET measurements were performed as a TRIBIC experiment. The ion beam was scanned over the circuit and the output of the circuit was connected to a Tektronix DPO72004 storage oscilloscope, that has a 20 GHz analog bandwidth and 50 G samples/s sampling rate. The circuit was mounted on a special board that had 50-X lines and the connections to the oscilloscope were made through 40 GHz cables and bias tees. When the oscilloscope was triggered (SET was observed), the beam was blanked at 100 ns rise time to allow the computer to transfer the transient waveform from the oscilloscope. The positions of the SETs and the complete waveforms were stored and processed off-line. In addition to the circuit level SET measurements, TRIBIC was performed on model transistors similar to those included in the circuit. The current transients on all four electrodes were recorded simultaneously with the ion beam position. 3. Results and discussion If we look at Fig. 1 we can see where most of the charge induction occurs. If we recall the Gunn theorem [8] it is obvious that the emitter and the base will have very little charge induced and the collector and substrate will produce the largest signal. For more detailed discussion of IBIC in this structure we refer the reader to [9]. The DTI reduces charge induction from hits outside of the body of the device. As shown in [9] it works very efficiently for ion hits outside of the deep trench. Recently, a question came up how this deep trench effects charge induction inside the deep trench. Fig. 2a and b shows the charge induction maps (collector signal) of the DT and non-dt devices while Fig. 3 shows the IBIC spectra for both devices. For a more detailed description of the devices see [10]. The first thing we should notice is that as we saw previously, the DT blocks very efficiently the charge from hits outside of the DT. There is a high charge induction area in the center (approximately the area of the device), then there is a narrow region where the charge induction drops to practically zero (this is where the deep trench is), and a smeared out region with circular symmetry where the induced charge is much less. On the other hand, in the non-dt device the structure is similar to the outside of the trench region of the T device. The surprising fact is that the induced charge in the non-dt device for direct hits (the ion goes through the device) is lower than in the DT device. This is very clear from Fig. 3 where the IBIC spectra are shown. We can qualitatively explain this phenomenon. When the ion hits outside of the DT device, the deep trench blocks the carriers diffusing into the subcollector collector junction. In the non-dt device these carriers can diffuse into the device area and induce charge when they cross the junction. On the other hand, the same DT also blocks the diffusion of carriers when the ion hits inside the deep trench. It means that the diffusion is blocked through the DT; therefore, the carriers can diffuse toward the junction more efficiently and finally induce more charge. In the non-dt device the carriers can diffuse laterally and recombine. Fig. 4a and b shows the simulated induced currents and charges for DT and non-dt devices using CFRDC s NanoTCAD [11]. The simulation was done for 7.5 6 lm 2 devices with center and outside (1 lm from the trench or its equivalent position on the non-dt device) hits. These figures clearly show that the reduction in the induced charge for center hits is due to the reduction of current in

2096 G. Vizkelethy et al. / Nuclear Instruments and Methods in Physics Research B 268 (2010) 2092 2098 the diffusion phase. The inset in Fig. 4b shows the induced current from outside hits. Notice that while the current scale for the active are hits is ma it is only la for the outside hits. The circuit diagram of the BGR circuit is shown in Fig. 5 marking the SiGe transistors of interest from Q1 toq5. Each transistor consists of four parallel 0.5 2.5 lm 2 HBTs, except Q2, which consists of 32 0.5 2.5 lm 2 HBTs. For a more detailed description of the circuit see [12]. Fig. 6 shows the physical layout of the HBT bank in the circuit, which is important to map the SETs. The RHDB design in order to decrease the induced charge (to reduce the amplitude of the SET) includes an n+-type implant surrounding the deep trench (n-ring). This n-ring creates a low impedance path to shunt electrons from the collector due to the reverse biased junction between the n-ring and p substrate, which creates a secondary electric field. A discussion of the different n-ring configurations and their effectiveness can be found in [13]. The top and cross-section views of both the standard and RHBD designs are shown in Fig. 7a and b (B base, E emitter, C collector, ST shallow trench, DT deep trench, and NR n- ring). Fig. 8a and b shows the location of the strikes over the HBT bank of standard and RHBD BGR circuits that resulted in SET at the output of the circuit. In both circuits, transistors Q5 and Q2 show the most sensitivity. Q3 is not sensitive at all and Fig. 9. SET waveforms on the output of the BGR hitting two different transistor structures (Q2 and Q5). Fig. 8. Maximum SET amplitude maps of the standard (a) and RHBD (b) devices after [12]. Fig. 10. Maximum current transient (TRIBIC) maps of standard (a) and RHBD (b) devices after [12].

G. Vizkelethy et al. / Nuclear Instruments and Methods in Physics Research B 268 (2010) 2092 2098 2097 area of the device, while the RHBD transistors have transient only for the emitter strikes. On the other hand the RHBD transistors show transient outside the transistors (although low amplitude) where the standard transistor had no transients. Fig. 11a and b shows transients from the active area hits for the standard (a) and for active area and outside hits for the RHBD transistors (b). As expected [12] the emitter hits produce large, very fast (<1 ns) transients. This is in agreement with the fact that this transient is mainly due to the drift of the carriers. For the RHBD device, the transient from outside of the device is smaller and lasts about five times longer than the active area hits. Also, notice that for an outside strike, the polarity of the collector current has changed which indicates that a different type of carrier is drifting toward the collector electrode. Based on the above results we think it is necessary to further study these devices using real mixed mode TCAD calculations. 4. Summary Fig. 11. Current transients (TRIBIC) from active area hits of for the standard (a) and RHBD (b) devices. All the electrode were at ground except the substrate that was held at 4 V. In the RHBD device the n-ring voltage was 2 V. although SET waveforms were recorded when Q4 was hit, they have small amplitudes. The large amplitude SETs (green stars) occur mainly when striking Q2 in the standard circuit and somewhat less frequently in the RHBD circuit. The RHBD circuit shows significantly reduced SET amplitude for strikes over Q2 and Q5, which indicates that the mitigation technique works to some extent. An unexpected result is that Q1, which was not sensitive at all in the standard design, produces low amplitude SETs in a circular pattern around the transistors. Fig. 9 shows the transients on the output of the circuit when Q2 (solid line) and Q5 (dashed line) are hit. The SET transients exhibit two peaks, one very sharp and a second one with a longer decay time, several hundreds of ns. We need to emphasize that these are not device responses but the captured responses at the output of the circuit. To further study the charge induction mechanism in this TRIBIC experiments on test structures of four standard/rhbd transistors were performed. Fig. 10a and b shows the maximum current amplitude maps (we need to emphasize here that these are real TRIBIC signals and not SET transients) for the standard and RHBD devices. Fig. 10a shows all four transistors of the standard devices while Fig. 10b shows only two of the RHBD transistors. In the latter case, the scan was zoomed in to see more details. Both maps show high amplitude current transients when the ion beam hits the emitter area of the transients. For strikes in the deep trench the standard transistors show transients over the whole We performed IBIC and SET experiments on SiGe HBT devices and circuits. We found that although the deep trench isolation reduces the induced charge for the ion hits outside of the deep trench, it increases the induced charge inside the deep trench by not letting the carriers diffuse outside. This can have significant impact on the design of new devices. SiGe bandgap voltage reference circuits were tested to study the SET response of SiGe analog circuits. We found that the RHBD devices (applying an n-ring around the transistor to reduce the induced charge) reduce somewhat the amplitude of the output transients but they introduce new sensitive areas that were not present in the standard design. We have to note that these new transients have small amplitudes. By testing individual test transistors we found that the emitter strikes showed quite similar response in both the standard device and the RHBD device, i.e., they had large amplitude and were very fast (1 ns). The RHBD device had no transient in the active area apart from the emitter in contrast to the standard device. But we captured current transients for outside of the active area strikes for the RHBD device with a pulse length of about five times of the emitter strikes. The SET transients from the circuit were found to be much longer (several hundreds of ns) than the transients from individual transistors, which indicates that circuits have more effect on the SETs than the individual devices. To further investigate the above phenomenon detailed mixed mode TCAD simulations have to be performed on these devices and circuits. References [1] J.D. Cressler, G. Niu, Silicon Germanium Heterojunction Bipolar Transistors, Artech House, Boston, London, 2003. [2] P.W. Marshall et al., Single event effects in circuit-hardened SiGe HBT logic at gigabit per second data rates, IEEE Transaction on Nuclear Science 47 (2000) 2669 2674. [3] R.A. Reed et al., in: Proceedings of IEEE Nuclear and Space Radiation Effect Conference of Data Workshop, 2001. [4] G.F. Niu et al., Simulation of SEE-induced charge collection in UHV/CVD SiGeHBTs, IEEE Transaction on Nuclear Science 47 (2000) 2682 2689. [5] A. Holmes-Siedle, L. Adams, Handbook of Radiation Effects, Oxford University Press Inc., New York, 2002. [6] M.B.H. Breese et al., A review of ion beam induced charge microscopy, Nuclear Instruments and Methods in Physics Research Section B Beam Interactions with Materials and Atoms 264 (2) (2007) 345 360. [7] G. Vizkelethy et al., Radiation effects microscopy for failure analysis of microelectronic devices, Nuclear Instruments and Methods in Physics Research Section B Beam Interactions with Materials and Atoms 231 (2005) 467 475. [8] J.B. Gunn, A general expression for electrostatic induction, its application to semiconductor devices, Solid-State Electronics 7 (1964) 739 742. [9] G. Vizkelethy et al., Ion beam induced charge (IBIC) studies of silicon germanium heterojunction bipolar transistors (HBTs), Nuclear Instruments and Methods in Physics Research Section B Beam Interactions with Materials and Atoms 260 (1) (2007) 264 269.

2098 G. Vizkelethy et al. / Nuclear Instruments and Methods in Physics Research B 268 (2010) 2092 2098 [10] S.D. Phillips et al., Impact of deep trench isolation on advanced SiGe HBT reliability in radiation environments, in: The Proceedings of the 2009 IEEE International Reliability Physics Symposium, Montreal, Canada, 2009. [11] NanoTCAD Software, Version 2008. Huntsville, AL, CFD Research Corp., September 2008. Available from: <http://www.cfdrc.com>. [12] L. Najafizadeh et al., Single event transient response of SiGe voltage references and its impact on the performance of analog and mixed-signal circuits, IEEE Transactions on Nuclear Science 56 (6) (2009) 3469 3476. [13] A.K. Sutton et al., An evaluation of transistor-layout RHBD techniques for SEE mitigation in SiGe HBTs, IEEE Transactions on Nuclear Science 54 (6) (2007) 2044 2052.