† Corresponding author. E-mail:
Project supported by the National Natural Science Foundation of China (Grant Nos. U1532261, 11690041, and 11675233).
Heavy-ion flux is an important experimental parameter in the ground based single event tests. The flux impact on a single event effect in different memory devices is analyzed by using GEANT4 and TCAD simulation methods. The transient radial track profile depends not only on the linear energy transfer (LET) of the incident ion, but also on the mass and energy of the ion. For the ions with the energies at the Bragg peaks, the radial charge distribution is wider when the ion LET is larger. The results extracted from the GEANT4 and TCAD simulations, together with detailed analysis of the device structure, are presented to demonstrate phenomena observed in the flux related experiment. The analysis shows that the flux effect conclusions drawn from the experiment are intrinsically connected and all indicate the mechanism that the flux effect stems from multiple ion-induced pulses functioning together and relies exquisitely on the specific response of the device.
Single event effects (SEEs) have become increasingly important as the feature size of the microelectronic devices continue scaling down.[1–3] Tremendous efforts, both experimental and theoretical, have been invested into this field to study the basic mechanism of the effects. Particle flux is an important experimental parameter in ground based irradiation tests. Actually, the natural space is a flux variable environment,[4,5] and the test standards have not specified the particle flux to be used in the ground based experiments.[6–8] In addition, the ion flux during SEE tests is normally much higher than that in real space radiation conditions. Thus, a dedicated study needs to take into account the flux impact on SEEs in microelectronic devices.
In the previous studies, Liu et al. introduced “double-hit” to explain the errors in the ADE harden device.[9] Edmonds suggested the flip of triple module redundancy (TMR) harden device is evidently raised at higher flux.[10] Yu et al. reported the test result differences between low and high fluxes for various types of devices and found that only certain types of devices exhibit flux dependence.[11] Our research group conducted the flux related experiment using various types of devices covering LET from 10.1 MeV·cm2/mg to 99.8 MeV·cm2/mg, and three experimental conclusions were drawn.[12] First, the flux effect is range dependent that memory devices become flux sensitive as the flux increases above 103 ions/cm2·s, i.e., for most cases, the cross section of the single event errors increases. Second, the flux effect is linear energy transfer (LET) dependent that as the LET of the incident ions becomes larger, the flux effect becomes more evident. Third, the flux effect is device dependent so different types of devices show different flux sensitivities.
This paper aims at studying the theoretical mechanism of the flux effect by employing GEANT4 and TCAD simulation methods. Radial charge distributions and the charge collection processes induced by various types of ions used in the ground based experiment are explored. The transient aspect of the heavy-ion track structure in silicon devices, together with the analysis of the specific device response, is investigated to explain the flux impact on SEEs in different memory devices.
GEANT4 is a powerful simulation tool kit utilizing the Monte Carlo method to calculate interactions of particles with matter.[13] It is programed in C++ language and widely used in high energy physics, nuclear physics, medical science, space science, etc. In this work, GEANT4 version 10.3 has been used by implementing the newly developed MicroElec model.[14,15] This new model is categorized in low energy electromagnetic physics especially developed for the interactions of particles in silicon by lowering the energy production threshold of secondary electrons down to 16.7 eV, which is much lower than that in the previous models. With the improvement of the new model, large numbers of discrete secondary electrons can be precisely tracked, giving a much better illustration of the ion track structure.
Three types of ions that are mostly used in the ground based experiments are chosen for the simulation: 209Bi, 129Xe, and 86Kr. In order to get sufficient statistics, the radial charge distribution for each type of ion is calculated by averaging the tracks of 106 incident ions.
The geometry used to simulate the radial charge distributions is illustrated in Fig.
The charge collection processes are simulated by the TCAD simulator Synopsys Sentaurus.[17] Figure
The first parameter to check is the LET spectra of the three different ions penetrating into silicon. The LET is calculated using the SRIM software and the LET curves of the three ions are depicted from the incident energy of 1 MeV to 2 GeV (Fig.
To investigate the track structure for the ion that has the same LET but different energy, the simulation is conducted using 129Xe ion at the LET of 53.7 MeV·cm2/mg with the energies of 130 MeV and 1677 MeV (Fig.
These two types of ions have the same LET at the surface of silicon that the energy deposition per depth is the same, but the radial profiles of the excited electrons are quite different. The average energy of secondary electrons for the 1677 MeV 129Xe ion is larger than that for the 130 MeV 129Xe ion. Hence, the energetic electrons excited by the 1677 MeV 129Xe ion can penetrate further away from the ion path and deposit their energies in larger distances. On the contrary, the average energy of secondary electrons excited by 130 MeV 129Xe is much lower, and the energy is deposited locally around the track core region.
The ions with the same LET but different energies produce remarkably different radial track structures in silicon. Figure
The radial charge difference results from the energy distribution difference of the δ-rays ejected by the incident ion. The simulation result of the radial charge distributions can be extrapolated to other ions with the same LET but different energies. On the right side of the Bragg peaks in Fig.
The incident energy of the ion plays an important role in determining the track radial profile. To compare the radial charge profiles of different ion spices at the energy of Bragg peaks, 209Bi, 129Xe, and 86Kr are simulated at the energies of 1020 MeV, 550 MeV, and 240 MeV, and the LETs are 99.8 MeV·cm2/mg, 69.1 MeV·cm2/mg, and 40.7 MeV·cm2/mg, respectively. Figure
The results concerning the charge collection processes extracted from the Synopsys Sentaurus simulations are reported in Fig.
The experiment results and analysis conclusions from the previous studies indicate that the flux effect might result from two or more ion-induced pulses affecting the storage cell at high flux radiation conditions. Hence, the spatial and temporal aspects of the ion-induced pulses are crucial to the mechanism of the flux related upset process. The results extracted from GEANT4 and TCAD simulations illustrate the charge deposition as well as the charge collection process. The charge collection length may be several micro-meters, and the charge collection time may last for several nano-seconds. If the charge sharing effect is considered,[20] the charge deposited at the strike location may be collected by nearby transistors, making the effective charge collection length even larger and the charge collection time even longer.
Furthermore, the particle incident in the device can cause significant voltage transients propagating in the circuit, affecting the sensitive nodes along the circuitry paths.[21] Thus, the affecting time by a single hit may be prolonged and the affecting area may be extended along the transients’ propagating paths in the circuit. Hence, the spatial and temporal requirement for the ion-induced pulses linking together is determined not only by the charge collection process, but more importantly, also by the detailed circuit response of the specific device.
As the flux continues to increase, the global radiation response of the devices will occur, which is termed as the dose-rate effect.[22–24] The transistors in the device are either connected to the power rail or to the ground bus. Each transistor behaves like the collection node under the global irradiation fluence, contributing to the current flow in the power rail or the ground bus. The induced currents across the entire chip can cause a diminished voltage supply, either a drop in Vdd and/or a rise in Vss, leading to a reduced noise margin and a lower transient radiation upset threshold. The final states of the cells are randomized by global circuit noises, and thus are more vulnerable to transient perturbations.
The experiment flux threshold of 103 ions/cm2 · s marks the boundary below which the flux effect will not happen. This might be explained by the fact that, when the flux is under 103 ions/cm2 · s, the opportunity of two ion-induced pulses linking together is relatively low and the dose-rate effect is negligible. The flux effect only occurs at the higher end of the flux range, i.e., when the flux increases above a certain value, the perturbations of multiple pulses coinciding together and the induced global current cannot be ignored. The boundary of 103 ions/cm2 · s does not mean that the flux effect will definitely happen when it is above this value, but denotes a strong possibility that the flux effect might happen. For certain type of device, the threshold for the flux effect might be even higher. When the flux continues to increase, there will be more radiation related, various types of internal noises complicating the upset mechanism, i.e., more transient pulses coinciding with other radiation induced noises contribute to the final cause of an SEU. Due to these trends, the error rate in memory parts tends to become higher when subjected to higher flux conditions.
This boundary value is related to the temporal cross sectional distribution of the ion beam, which has something to do with the beam transportation and modulation mechanism of the high energy ion acceleration system. A different accelerator facility may have a different accelerating and transportation mechanism, leading to a different transient cross sectional distribution of the ion beam, and thus a different boundary value for the flux effect.
As the LET becomes larger, two aspects of the radial charge profile might be changed: the charge distribution radius and the charge deposition density. From the above analysis about the flux effect, the charge collection length in the device and the charge collection amount by the sensitive node can both contribute to the flux effect, yet in quite different ways. As the charge collection length becomes larger, the chance of direct special connection becomes larger, making the storage cell more likely to be impacted by two or more ions. As the deposited charge density becomes larger, the collected charge by the sensitive node becomes larger, making the disturbing pulse larger and the dose rate effect more significant.
The experiment conclusion is drawn based on the data taken at LETs near the Bragg peaks. Figure
The simulation result suggests that for the ions at the same LET, the radial track profiles are different when the ion masses and energies are different. The charge distribution radius is larger when the ion energy is larger, and the charge density near the track core region is larger when the ion energy is smaller. Hence, the flux effect is not only related to the LET of the incident ion, but also to the ion mass and energy. The question regarding which aspect is more important to the flux effect needs further experiments to verify.
In our previous study, hardened and unhardened devices were subjected to the flux related irradiation experiment. The hardened devices include SOI SRAMs hardened by ADE, bulk-Si flip-flops hardened by dual interlocked storage cell (DICE), and triple modular redundancy (TMR). The test result is consistent with the others that the hardened devices are more flux sensitive than the unhardened ones. As the flux increases, the error rates of the hardened devices are evidently raised.
The most sensitive strike location in the SRAM cell for most technologies is the off n-channeled transistor and off p-channeled transistor.[25] Figure
Figure
Figure
Single particle hitting a single sensitive node is not likely to cause an upset in the hardened cell. Redundant nodes restore the state of this affected node and prevent an upset in the storage cell. However, as the flux increases above 103 ions/cm2 · s, the noise margin might be reduced and there might be two or more ion-induced pulses affecting separated sensitive nodes during the same operation cycle of the storage cell. With multiple ion-induced pulses disturbing the separated sensitive region, it may result in the flip of the hardened cells.
The flux effect is also evidenced from the failure mode of the unhardened bulk-Si SRAM. As the flux increases, the overall cross section of the unhardened SRAM deviates with no specific trend, but the MBUs of more than 3 bits increase as the ion flux increases.[26,27] At the flux higher than 103 ions/cm2 · s, four or even higher bits errors appear that must result from more than one ion hitting the adjacent region during the same time interval.
The SEU cross section is denoted by
GEANT4 and TCAD simulations are performed to analyze the radial charge profile and charge collection time for three different ions mostly used in the ground based experiment. The radial track structure is determined not only by the LET, but also by the mass and energy of the incident ion. The overall charge collection time for different types of ions may last for nano-seconds. The simulation results are analyzed to explain the flux effect, that only at the higher end of the range, higher than 103 ions/cm2 · s, can the flux effect occur. The flux effect is related to the amount of collected charge and the size of collection area, therefore the LET and the energy of the incident ion. Different device types are investigated to explain the different flux effect and the reason for the SEE cross section increment is analyzed.
The flux of 103 ions/(cm2 · s) is recommended for the ground-based test. As the flux continues to increase, the global radiation response related to the flux effect will occur, making all hardening methods fail. For the flux increases above 105 ions/(cm2 · s), the flux effect will be even worse and flux related other perturbations will happen, which is highly unlikely in the real space irradiation condition. However, it provides a method to test whether the device has hardened enough.
The analysis of the mechanism involving the flux effect has a significant impact on designing radiation-hardened circuits at the smaller technology scales. The suggestion is consistent with the test standards that the flux higher than the 105 ions/cm2 · s is not recommended. It will play an important role in the selection of test conditions to limit the necessary testing time.
[1] | |
[2] | |
[3] | |
[4] | |
[5] | |
[6] | |
[7] | |
[8] | |
[9] | |
[10] | |
[11] | |
[12] | |
[13] | |
[14] | |
[15] | |
[16] | |
[17] | |
[18] | |
[19] | |
[20] | |
[21] | |
[22] | |
[23] | |
[24] | |
[25] | |
[26] | |
[27] |