Devices under test and test system

The 28 nm Kintex-7 FPGAs from Xilinx were chosen as the devices under test (DUTs). The Kintex-7 family of FPGAs offers substantial programmable logical configuration resources (that is, CRAMs). The Xilinx 7 series devices operate on the VCCINT power supply rail, and the voltage ranges from a minimum of 0.9 V to a maximum of 1.0 V. To meet the range requirement of the radiation tests, the substrate was thinned to approximately 50 μm.

To better characterize the various test vectors in radiation experiments, a test board was developed, and the diagram is illustrated in Fig. 1. The entire test system consists mainly of the motherboard and daughterboard. A Kintex-7 FPGA obtained from Xilinx was selected and placed in the motherboard to control the configuration, readback, and comparison of the DUT configuration. The bitstream used to configure the DUT is stored in an external flash. CRAM can be characterized in static or dynamic modes. The control modules help to read the bitstream back. The error comparison between the radiation and golden bitstreams is performed in the XC7K325T chip. In addition, during the radiation tests, the error data and information were transmitted to a personal computer through an RS 232/485 converter. The separate structure of the test system makes it possible to perform arbitrary hardware designs to satisfy diverse testing requirements.

Fig. 1Full-screen View

Diagram of the radiation test system

The monitor modules were designed to read and record both the voltage and current information of the DUT during the radiation tests. The power modules are responsible for powering the entire test system and maintaining a stable voltage. In addition, a C#-based customized software with a visual interface was designed to make the control and data analysis more accurate and convenient. This allowed for the observation and recording of the radiation data in real time. During the experiment, the computer placed in the test room was connected to our test board to control the test system by operating the software.

Experimental information

High-energy ⁷⁸Kr and medium-energy ¹²⁹Xe heavy ion experiments were performed at the Heavy Ion Research Facility in Lanzhou (HIRFL) at the Institute of Modern Physics, Chinese Academy of Sciences. Irradiation tests were conducted in air and at a nominal temperature. The ⁷⁸Kr and ¹²⁹Xe ions with initial energy of 60 and 19.5 MeV/u were provided by high-energy radiation effect (HERE) terminal and terminal No.5 (TR5), respectively. For tests at TR5, the devices must be thinned down to ensure that heavy ions reach the sensitive region inside the DUT. For high-energy tests, the HERE terminal in HIRFL can accelerate heavy ions to hundreds of MeV per atomic mass unit, which provides a wide energy spectrum for SEU investigations. In addition, the heavy ions provided by the HERE terminal have a very long range. Thus, the prototype devices can be easily evaluated without pre-thinning or de-capping.

In the experiments, aluminum degraders of different thicknesses were added in front of the DUT to obtain several appropriate LETs. The DUT radiated in TR5 was fixed on a 5-dimensional platform to change the positions in three directions and to rotate at two angles (azimuth θ and zenith ψ), as shown in Fig. 2. Static modes were employed in both the high-energy and medium-energy tests. This prevents unexpected errors, such as burst errors during reconfiguration. Table 1 lists the detailed information used in the experiments. Multiple tilts and voltages were coupled to explore the radiation sensitivities of the DUT, which were helpful in obtaining the worst MBU responses. Prior to the tests, a specific circuit was configured by loading a known bitstream into the CRAM. After each trial, the configuration bits were read back via the SelectMAP interface and compared with the golden bitstreams to distinguish the upsets. Finally, log files with the number of errors and logical locations were generated to identify the MBU patterns.

Fig. 2Full-screen View

(Color online) Diagram of the radiation angles: azimuth θ and zenith ψ

HEHI tests

Figure 3 graphs the cross sections of HEHI tests under different coupling parameters. For three LETs, the cross sections exhibited good consistency. Under all conditions, 0.9 V tests had higher cross sections than those of 1.0 V. This is because heavy ion tracks in the physical layout can cover more 0.9 V cells than 1.0 V to create larger sizes and higher percentages of MBUs. The increased cross sections from 1.0 to 0.9 V proved that the critical charge reduction caused by the decreased voltages requires close attention from SEU evaluators. Furthermore, the large tilt dependence of the SEU cross sections was identified. Compared with voltage, the influence of tilt was remarkably obvious, which is critical for the MBU evaluation of FPGAs. The maximum discrepancy in cross sections induced by tilted radiation can reach approximately 43.8%, revealing that when HEHIs with a higher LET are tilted incident into FPGAs, more cells would be affected to create more MBUs. Furthermore, compared with incidence under the vertical condition, the sharp increase of the cross section under the (0°, 30°) condition from the LET of 20.8 to 31.0 MeV/(mg/cm²) indicated that ⁷⁸Kr ions with stronger ionization ability can cause more serious MBUs. In actual space, these MBUs would significantly threaten the reliability of FPGA-based on-orbit systems, which is important in radiation evaluations. When the combined effect of both the angle and bias is considered and the LET is equal to 31.0 MeV/(mg/cm²), the discrepancy reached 46.2%, as shown in Fig. 3 (the difference between the red dot and blue triangle). Additionally, since the actual space environment has worse conditions, the coupled parameters are extremely necessary during radiation evaluation to prevent the overestimation of SEU tolerance.

Fig. 3Full-screen View

Cross sections of CRAM vs. LET under different angles and voltages

Medium-energy heavy ion tests

The cross sections of the CRAM irradiated by ¹²⁹Xe ions, shown in Fig. 4, demonstrated good consistency as a function of the angles. Additionally, the upsets exhibited different dimensional dependences. In particular, the cross section of (0°, 30°) was higher than that of (90°, 30°). During the analysis, we found that, for the vertical radiation condition, upsets mainly occurred along the word direction rather than between words. Furthermore, larger MBUs appeared when performing the (0°, 30°) tests. However, in the (90°, 30°) and (90°, 45°) conditions, multiple events occurred mainly between adjacent words. The above results confirmed the shorter distances between two adjacent cells along the word direction. Furthermore, it revealed that the different physical distances between sensitive nodes in different directions significantly affect the device sensitivity. For the DUT used in this study, the closer distance between cells in the word direction should be given special attention by radiation evaluators. In addition, radiations under ψ=30° had more serious impacts on the FPGA function, with a maximum of approximately 1.7× increment from (90°, 0°) to (45°, 30°).

Fig. 4Full-screen View

Total CRAM SEU cross sections induced by ¹²⁹Xe ions at different tilts

MBU results of HEHIs

Table 2 shows the percentages of SBUs, 2-bit MBUs, and greater-than-2-bit MBUs under all experimental conditions. When the LET was 13.3 MeV/(mg/cm²), as the small LET of HEHIs, the influence of the angle on the percentage of MBUs was not obvious. However, as the LET increased, the tilt-induced percentages of 2-bit MBUs significantly increased, which explains the sharp increase observed in Fig. 3. Owing to the decreased critical charge, more MBUs appeared during the 0.9 V operation.

To further reveal the influence of tilt and voltage on the scales and percentages of MBUs, detailed information of greater-than-2-bit MBUs was extracted, as shown in Fig. 5. As mentioned in Sect. 3.1, the 0.9 V radiations induced higher cross sections. Although the increase was small, they had a greater influence on the MBU sizes. Compared with the 1.0 V, the 0.9 V radiations induced a higher proportion of and larger scale MBUs. At normal incidence of 13.3 MeV/(mg/cm²), due to the small LET of HEHIs, only 0.9 V radiations induced 5-bit MBUs, whereas 1.0 V radiations did not. When the LET was 31.0 MeV/(mg/cm²), the (0°, 30°) condition tilted ⁷⁸Kr ions induced larger 6-bit MBUs in the 0.9 V DUT than in the 1.0 V, which is the same as the normal incidence of ⁷⁸Kr ions with 13.3 MeV/(mg/cm²). We speculated that this is because these two heavy ions with different energies have different ionization-track structures. In tilted radiation, the lowered 0.9 voltage compared to 1.0, can produce more MBUs. Additionally, we found that compared with 13.3 and 31.0 MeV/(mg/cm²), ⁷⁸Kr ions with an LET of 20.8 MeV/(mg/cm²) can cause higher percentages of greater-than-3-bit MBUs in vertical radiations at 0.9 and 1.0 V, and greater-than-4-bit MBUs in tilted radiations at the 0.9 V operation. This is related to the different influence ranges and abilities of the ionization tracks.

Fig. 5Full-screen View

(Color online) Detailed percentages of greater-than-2-bit MBUs of different types at different conditions

Because the (0°, 30°) ion trajectories covered more cells, the largest 6-bit MBUs appeared even at the LET of 13.3 MeV/(mg/cm²), which identified the worst influences in this study. Compared with the LET impact, the tilt played a more important role in the generation of larger MBUs and high percentages of greater-than-3-bit MBUs, which can be determined from the comparison both between 0.9 V-13.3 MeV/(mg/cm²)-(0°, 30°) and 0.9 V-31.0 MeV/(mg/cm²)-(0°, 0°), and 1.0 V-13.3 MeV/(mg/cm²)-(0°, 30°) and 1.0 V-31.0 MeV/(mg/cm²)-(0°, 0°). The maximum 6-bit MBUs all appeared under the condition of (0°, 30°) incidence. The more severe impacts of the ion trajectories revealed that the tilted ⁷⁸Kr ions had a wider track influence range on the layout of the DUT.

In the comparison of two different voltage radiations, at LET equal to 31.0 MeV/(mg/cm²), the increased 2-bit MBUs appeared to cause the roll up in the overall MBU percentages. However, under the low LET conditions of 13.3 and 20.8 MeV/(mg/cm²), the higher percentages of MBUs at 0.9 V were mainly due to the generation of greater-than-2-bit MBUs whether at normal or tilted incidence. These results indicated that different track structures of heavy ions with different energies can induce different MBU types. For all the results illustrated in this subsection, the reduced voltages and tilted HEHIs showed worse MBU effects, which clearly demonstrated that the characterization of large-scale MBUs caused by worst parameters is indispensable in radiation evaluations.

MBU results of ¹²⁹Xe ions

Figure 6 shows the percentages of SBUs, 2-bit MBUs, and greater-than-2-bit MBUs caused by ¹²⁹Xe ions at various tilts. With increasing tilt, the percentage of MBUs also increased. More MBUs appeared at (0°, 30°) than at (90°, 30°); that is, more obvious dimensional dependence was identified. The significant MBU orientation effects implied that the incident directions of ions with respect to the layout of the cells are crucial. For the greater-than-2-bit MBUs, the maximum approximately 2.87× increment between (90°, 0°) and (45°, 30°) made the functionality of our customized design abnormal.

Fig. 6Full-screen View

(Color online) Percentages of SBUs, 2-bit, and greater-than-2-bit MBUs under different angles

To further reveal the effects of different dimensions on the MBU scales and percentages, the detailed results of greater-than-2-bit MBUs are shown in Fig. 7. Only the largest 3-bit MBUs were found at (90°, 0°). When setting θ=90° while rotating ψ, we captured MBUs no larger than 4-bit; however, for ψ=30° while changing θ, the largest 5-bit MBUs appeared six times, which helped us determine worse dimensional impacts. In addition, the percentages of 3-bit MBUs at (0°, 30°) were approximately twice as large as those at (90°, 30°), further demonstrating that there are indeed different sensitivities in the directions of either the word or between words in our DUT. We observed that large-scale upsets occurred between words in our DUT at θ=90°, but the largest number was four. When considering the largest 5-bit MBUs, we believed that the distances between word cells are closer than the cells between words.

Fig. 7Full-screen View

Detailed percentages of greater-than-2-bit MBUs at various tilts

Geant4 simulation

When comparing the MBU results between ⁷⁸Kr and ¹²⁹Xe ions, it was found that, although HEHIs have a much lower LET, they can induce even larger MBUs. This appears to be caused by the different ionization track structures. Hence, using the Geant4 toolkit, we performed Monte Carlo simulations to investigate the track structures of the ⁷⁸Kr and ¹²⁹Xe ions used in our experiments. The results are shown in Fig. 8. It can be seen that HEHIs have much wider tracks; particularly, the ⁷⁸Kr ions of 3348 MeV were approximately five times wider than the ¹²⁹Xe ions. Large clustered MBUs occurred due to their large track radius. However, owing to their relatively low density of electron-hole pairs inside tracks, the largest 6-bit MBUs appeared only under the (0°, 30°) tests. The tilted incidence would cause a widened track to affect more cells, indicating that the larger track area of HEHIs is more catastrophic in the generation of larger MBUs than ¹²⁹Xe ions with a much higher LET.

Fig. 8Full-screen View

Electron-hole pairs density changing as a function of the radial distance in silicon

As mentioned in Sect. 4.1, we found that the ⁷⁸Kr ions of 1653 MeV had a more serious impact on the MBU percentages. Fig. 8 displays these results well. Although the ⁷⁸Kr ions of 1653 MeV have a similar ionization radius to those of 3348 MeV, they generated more electron-hole pairs during their interaction with transistors, leading to a higher probability of MBUs. In addition, in contrast to the ⁷⁸Kr ions with 793 MeV, although the ⁷⁸Kr ions of 1653 MeV have smaller electron-hole pair densities along their trajectories, their wider ionization tracks could result in more sensitive drains of transistors collecting charges to be probabilistically upset. That is, as the ⁷⁸Kr ions with an energy of 1653 MeV have a better trade-off between the wide ionization track radius and electron-hole pair density, more MBUs occur.

In addition, the track structures well explained the high-energy results observed in the experiments. When the LET was relatively small (that is, 13.26 and 20.83 MeV/(mg/cm²)), the 0.9 V radiations induced mainly larger scale MBUs than at 1.0 V; whereas at 31.04 MeV/(mg/cm²), HEHIs caused more 2-bit MBUs. For the almost equal ion track radius of HEHIs of 1653 and 3348 MeV, as ⁷⁸Kr ions with lower energy can deposit energy in more cells, more 0.9 V transistors may be upset at the edge of the track that exceeds the critical charge of the 28 nm FPGA cells. Thus, the generation of greater-than-2-bit MBUs led to an increase in the overall MBU ratio. However, despite the higher density of ionized charge at the center of the track for ⁷⁸Kr ions of 793 MeV, the narrower track made the 2-bit MBUs responsible for the increased percentage of MBUs induced by 0.9 V.

The heavy-ion-induced ionization track is shaped like a cylinder and can be up to several microns in diameter in silicon. However, tracks may sometimes be deformed depending on the selection of experimental parameters, such as angles. A diagram of the relationship between the ionization tracks and memory cells is shown in Fig. 9. In nanoscale devices, the heavy-ion track usually covers several memory cells. For heavy ions with a given energy, the vertical view of the ionization track structure is roughly circular; however, when conducting tilted radiation tests, the track changes to an ellipse, as illustrated in Fig. 9(b). Taking the MBU orientation differences in the sensitivities observed in our radiation results as an example, the widened major axis of the elliptical track deposits energy in more cells. Although these cells collect relatively less charge than those closer to the center of the track, the collected charge may still cause upsets in modern advanced devices with small critical charges. Moreover, if the major axis of the ellipse is oriented in the word cells, as in our Kintex-7 FPGAs, then larger MBUs would appear, which are difficult to correct using error mitigation (such as error-correcting code). In both ⁷⁸Kr and ¹²⁹Xe ion radiation, because we performed two different dimensional tests, as shown in Fig. 9, we observed that the broadened track induced larger clustered events along and between words. Their orientations depended on the degree of θ used. In this study, radiation at θ=0° can cause more and larger MBUs; therefore, the word direction of our DUT was confirmed. As θ=90° orients the direction between words, the MBU percentages were almost equal for the three conditions we used in the ¹²⁹Xe ion tests. The widened tracks of HEHIs under tilted conditions significantly affected the MBU characterization in advanced integrated circuits.

Fig. 9Full-screen View

(Color online) Diagram of the relationship between heavy ion tracks and memory cells for (a) vertical, and (b) tilted heavy ion radiation

As advanced nanoscale commercial FPGAs have small critical charge, the wider tracks of HEHIs have worse effects than medium-energy heavy ions on large scale MBUs. In particular, when performing tilted radiation, the reliability of the devices will be more seriously threatened. As essential factors in the actual space environment, more severe MBUs caused by the reduced voltage should also be carefully considered. The above results for both high-energy and medium-energy heavy ions showed that comprehensive considerations of actual space factors will be closer to the device evaluation requirements. Thus, to fully characterize the MBU features and ensure the successful implementation of effective MBU-hardened designs, it is necessary to take HEHIs and these comprehensive radiation conditions into account when designing ground-based accelerator tests.