2.1 Proton irradiation setup

The protons were produced by a cyclotron accelerator and adjusted by a quadrupole magnet, slit, and scatter rings before reaching the terminal. The proton irradiation terminal at the CIAE is shown in Fig.1. At this terminal, the proton beam was processed by a depletion film and collimator to hit the device under test (DUT). The DUTs were mounted on a sample holder, which was movable. During the experiments, various areas of the DUT were irradiated by moving the sample holder without fluctuating the beam spot. The beam spot size ranged from 1 cm × 1 cm to 10 cm × 10 cm. The flux of the proton beam could be adjusted from 10⁵ p·cm^-2·s^-1 to 10¹² p·cm^-2·s^-1 [8].

Fig.1.Full-screen View

The proton irradiation terminal at the China Institute of Atomic Energy.

As Fig.2 shows, the host computer communicated with the DUT via a universal asynchronous receiver/transmitter interface and programmable power was supplied to the DUT through a Bayonet Neill–Concelman cable. The power was kept at 5 V and monitored for possible SELs. The experiments were performed at room temperature and the chip temperatures were also recorded in real-time by an integrated temperature sensor.

Fig.2.Full-screen View

The schematic of the irradiation test at the China Institute of Atomic Energy.

2.2 Single-event effect test

The DUT was a Xilinx Zynq-7000 28 nm SoC [17] and the OCM block was tested in the experiments. The basic structure of the OCM was comprised of 6 transistors and the area of one cell was about 650 nm × 250 nm. The size of the chip was about 1.8 cm × 1.8 cm, and the entire chip was irradiated during the tests. Two boards were mounted on the sample holder. One was irradiated with 90 MeV protons and the other with 70 MeV protons.

For the OCM test, 64 kB of data was tested dynamically. The test pattern data, 0xA5A5A5A5, was written into the addresses and read back continuously. The SoC compared the read-back data with the expected data to detect single-event upset (SEU) occurrences. Once an event occurred, the information, including the failure addresses and upset bits, was recorded in text files and displayed on the HyperTerminal simultaneously.

For the 90 MeV proton irradiation, the flux was about 1.3 × 10⁸ p·cm^-2·s^-1, and the flux for the 70 MeV proton irradiation was about 2.3 × 10⁸ p·cm^-2·s^-1. The cumulative fluence was 1.0 × 10¹¹ p·cm^-2 for both irradiation energies.

3.1 Irradiation results

Both SEU and single-event functional interruption (SEFI) events were detected in the two irradiation tests, and no SELs were captured. There was no obvious drift in the chip’s currents and temperatures during the irradiation. In the 90 MeV test, 143 SEU and 7 SEFI events were detected. In the 70 MeV test, 118 SEU and 6 SEFI events were detected. The SEU events consisted of SBUs and multi-cell upsets (MCUs).

The results of the 90 MeV test are shown in Table I, including SBUs, 2-cell upsets, 3-cell upsets, 4-cell upsets, 5-cell upsets, and the uppermost 9-cell upsets. SBUs and 2-cell upsets dominate the results. Although 4-cells upsets, 5-cell upsets, and 9-cell upsets were discovered only once, this indeed demonstrates that the 28 nm SoC is very sensitive. In terms of the SBUs in the 90 MeV test, the ratios of 0→1 upset and 1→0 upset were 52.9% and 47.1%, respectively. For the 2-cell upset events, the ratios of 0→1 upset and 1→0 upset were 44.4% and 55.6%, respectively. These phenomena indicate that the probabilities of 0→1 upset and 1→0 upset were similar for the unhardened SoC.

The results of the 70 MeV test are presented in Table II. Similar to the results of the 90 MeV test, the majority of the events were SBUs and 2-cell upsets, with an uppermost number of 4 upset cells. It can be seen that the proportions of 0→1 upset and 1→0 upset were 52.2% and 47.8% respectively for SBUs and 55.6% and 44.4% respectively for 2-cell upsets. This verifies that the probabilities of 0→1 upset and 1→0 upset were again similar for the unhardened SoC. These results are consistent with the non-hardened 6-transistor memory irradiation test reported in [18].

The SEU bit cross section can be obtained from Eq.(1), where σis the cross section in cm²·bit^-1, n is the upset number, Ф is the cumulative fluence in cm^-2, and N_bit is the total number of tested bits. The device cross section can be calculated without N_bit in Eq.(1) and in cm².

For the 90 MeV and 70 MeV proton irradiation tests, the device cross sections of the SBUs were 1.02 × 10^-9 cm² and 0.88 × 10^-9 cm², respectively. Moreover, the bit cross sections were (1.95 ± 0.39) × 10^-15 cm²·bit^-1 and (1.67 ± 0.37) × 10^-15 cm²·bit^-1, respectively.

In the experiments, the striking energies of the protons were 90 MeV and 70 MeV, and their effects depended more on the secondary particles depositing energy; because the proton energies and their mechanisms were similar, their cross sections were therefore also similar.

As reported in [16], an OCM was tested with 18 MeV protons and no other effects were detected except for SBU events. However, in our work, the detected events contained several types, including SBU, MCU, and SEFI. The SBU cross section reported in [16] is about 8.0×10^-15 cm²·bit^-1. There are some discrepancies with our results, although they are of the same orders of magnitude. The equivalent LETs and the 50% and 90% critical radii for protons of differing energies in silicon have been simulated by Geant4 [19]. When compared with 18 MeV protons, the equivalent LETs for 70 MeV and 90 MeV protons are larger and the 50% and 90% critical radii are longer. This illustrates that it is possible for 70 MeV and 90 MeV protons to induce MCU events. For 70 MeV and 90 MeV protons, these parameters are approximate, and that is why the SBU cross sections of the 70 MeV and 90 MeV protons are similar.

The cross sections of the 2-cell upsets and MCUs are shown in Fig.3. For 2-cell upsets and 3-cell upsets, the cross sections of the 90 MeV and 70 MeV protons are similar. For 4-cell upsets, 5-cell upsets, and even 9-cell upsets, there is a difference between the 90 MeV and 70 MeV protons owing to the occurrence of very few upset numbers. Additionally, the SEFI cross sections of the 90 MeV and 70 MeV proton tests were calculated and were approximately 7 × 10^-11 cm² and 6 × 10^-11 cm², respectively.

Fig.3.Full-screen View

The cross sections of the 2-cell upsets and multi-cell upsets from the 90 MeV and 70 MeV proton tests.

Consequently, we can make the preliminary conclusion that the effects of 90 MeV and 70 MeV protons on a 28 nm SoC are similar.

3.2 Simulation results

To analyze the experimental results further, Cosmic Ray Effects on Micro-Electronics Monte Carlo (CRÈME-MC) simulations were performed [20, 21]. For the CRÈME-MC simulations, a vertical sketch of the chip was created based on reverse engineering and extracted information. The thicknesses of the layers are listed in Fig.4. The size of the module was 0.7 µm × 0.3 µm × 19.78 µm, and the number of incident particles was 10⁷. The sensitive volume was 160 nm × 160 nm × 160 nm and the critical charge was 0.18 fC.

Fig.4.Full-screen View

The vertical sketch of the 28 nm SoC used in the Cosmic Ray Effects on Micro-Electronics Monte Carlo simulations, created with information extracted from reverse engineering.

As Fig.5 shows, the cross-section results of the CRÈME-MC simulations for the 90 MeV and 70 MeV protons were 1.68 × 10^-15 cm²·bit^-1 and 1.41 × 10^-15 cm²·bit^-1, respectively. Compared with the SBU cross sections obtained from the experiment, the CRÈME-MC simulations reported in this paper are convincing. However, the simulations demonstrate the trend that the cross sections of lower energies were higher than those of higher energies by 4-5 orders of magnitude. Further, the peak values were obtained at 2 MeV. Effects could be observed only when the proton energies were greater than 1.4 MeV. For 1.4 MeV protons, the effects were caused by direct ionization, the corresponding LET was 0.142 MeV·cm²·mg^-1, and the range in silicon was 27.22 µm [22].

Fig.5.Full-screen View

The results of the Cosmic Ray Effects on Micro-Electronics Monte Carlo (CRÈME-MC) simulations.

3.3 Results discussion

The numbers of SEEs for the 90 MeV and 70 MeV proton tests were 150 and 124, respectively. These numbers are low when compared with the number of tested bits, which was 524288. The deviations of the tests can be obtained by Eq.(2). In this formula, α is the significance level, ${\varnothing }^{-1}(\frac{\alpha }{2}$) is the inverse cumulative standard normal distribution function with a corresponding confidence level of (1-$\alpha )$, F is the number of detected errors, and N is the total number of tested bits, and ɛ% is the bounds of true probability. As shown in [23], the estimated error probabilities were 2.86 × 10^-4 and 2.36 × 10^-4 for the 90 MeV and 70 MeV proton tests, and the deviations were about 16.01% and 17.6% with a 95% confidence level, respectively.

The cumulative dose can be calculated from the SEE test with an accuracy of ±10% by Eq.(3), where D is the deposited dose in rad, Ф is the fluence in cm^-2, and LET is expressed in MeV·cm²·mg^-1 [24]. In the test, Ф was 10¹¹ p·cm^-2 and the LETs of the 90 MeV and 70 MeV protons were 0.00632 MeV·cm²·mg^-1 and 0.0076 MeV·cm²·mg^-1, respectively. The total doses were 10.11 krad and 12.16 krad, respectively. These results indicate that the tolerance dose for the device was possibly higher. This is in agreement with the results reported in [13]. The current range during the test was 0.35-0.37 A for the two boards without spiking.

As shown in Fig.5, the CRÈME-MC simulation results are consistent with the experimental results. Moreover, the 18 MeV proton SBU cross section based on the CRÈME-MC simulation was about 6.5 × 10^-15 cm²·bit^-1, which is also consistent with the results reported in [16]. This demonstrates that the constructed CRÈME-MC module shown in Fig.4, including a module size of 0.7 µm × 0.3 µm × 19.78 µm based on the extracted vertical structure, a sensitive volume of 160 nm × 160 nm × 160 nm, and a critical charge of 0.18 fC, can be used to preliminarily simulate the SEE sensitivity of the SoC and that the results can be analyzed further.

Fig.5 shows that the cross sections of lower energies (<4 MeV) were higher than those of higher energies (>10 MeV) by 4-5 orders of magnitude. This implies that factors from nuclear reaction to direct ionization cause upset in a 28 nm SoC. As reported in [25], proton-induced SEEs in a 90 nm static random-access memory (SRAM) were analyzed. In [26], the proton SEE data for a 65 nm SRAM were presented. As Fig.6 shows, the SEE cross sections of low-energy protons are higher than those of high-energy protons by multiple orders of magnitudes. As the technology scales down, the peak cross section moves in the higher-energy direction: 0.7 MeV for a 90 nm SRAM [25], 1.2 MeV for a 65 nm SRAM [26], and 2 MeV for a 28 nm SRAM. The reason behind this observation is that as the thickness of the passive layer increases, higher energies are stopped. Specifically, the thickness of the passive layer is less than 5 µm for the 90 nm SRAM reported in [25], about 5 µm for the 65 nm SRAM reported in [26], and more than 10 µm for the 28 nm SRAM in this work. Moreover, in silicon, the energy at which proton inelastic scattering reactions begin is about 3 MeV. A cross section valley may appear in the interval between the energy of the peak SEU cross section and 3-4 MeV (the reaction cross section is very small at the beginning). As shown in Fig.6, for a 90 nm SRAM, a cross section valley appears at approximately 4 MeV. As the size of the manufactured technology decreases, the energy of the peak cross section shifts to the right, the critical charge decreases, and there is no cross section valley for smaller-scale SRAM. In a 28 nm SRAM, direct ionization causes effects at low energies. The energy interval for direct ionization is slightly wide owing to its low critical charge. This can lead to a direct cross-over of the interval in which there is a low probability of reaction. The consequence of this is that there will no longer be a cross section valley for the 28 nm SRAM.

Fig.6.Full-screen View

Proton-induced single-event effects in the various static random-access memory cell technologies.

To analyze the similarities of the effects of the 90 MeV and 70 MeV protons, and to further explain the differences between this work and that reported in [16], a Geant4 simulation of protons striking silicon was performed [27, 28]. In the simulation, the size of the silicon detector was 7 µm × 3 µm × 20 µm, and the particle number was 5 × 10⁶. Because the effects of medium-energy protons are mainly produced by nuclear reactions, the secondary charged heavy ions from the nuclear reactions were analyzed for 18 MeV, 70 MeV, and 90 MeV protons, respectively.

Table III lists the majority of the secondary charged heavy ions from the 18 MeV, 70 MeV, and 90 MeV protons hitting the silicon detector. It can be seen that ²⁸Si and ²⁷Al are the major secondary particles from the 18 MeV protons, and ⁴He, ²³Na, ²⁴Mg, ²⁷Al, ²⁷Si, and ²⁰Ne are the main secondary particles from the 70 MeV and 90 MeV protons. The secondary particles from the 70 MeV and 90 MeV protons are obviously more numerous in type than those from the 18 MeV protons. The LET intervals of the secondary particles from the 70 MeV and 90 MeV protons can be much broader than those of the secondary particles from the 18 MeV protons. This can lead to a difference in the observed effects between the 70 MeV and 90 MeV protons and 18 MeV protons.

The LET intervals of the various particles are depicted in Fig.7 in which, for example, 18-Si means that Si is the secondary particle from an 18 MeV proton interacting with silicon, and the other annotations have similar meanings. As shown in Fig.7, except for Si, the LETs of the other particles from the 90 MeV and 70 MeV protons are similar, and this can explain why the effects of the 90 MeV and 70 MeV protons are similar but their uppermost numbers of upset cells are different.

Fig.7.Full-screen View

The linear energy transfer (LET) intervals of the various secondary charged heavy ions in this work.

Similar to Fig.7, the ranges of the intervals in silicon for the secondary particles are shown in Fig.8, in which (a) displays the ranges of the other particles and (b) depicts the ranges of He. As can be seen in this figure, the ranges of the particles from the 90 MeV protons are generally larger than those of the particles from the 70 MeV protons. It is more likely that 90 MeV protons will induce a greater number of upset cells than 70 MeV protons.

Fig.8.Full-screen View

The ranges of the various charged heavy ions in this work. (a) The ranges of the other particles. (b) The ranges of He.

After comprehensively analyzing the results of our work, the cross sections reported in [16], and the results of the CRÈME-MC and Geant4 simulations, we can conclude that the effects of the 90 MeV and 70 MeV protons were similar because the majority of the secondary charged heavy ions produced by the two proton energies and the values of their LETs were similar. In addition, a greater number of MCUs was possible owing to 90 MeV protons than 70 MeV protons because the ranges of the secondary particles from the 90 MeV protons in silicon were larger than those of the secondary particles from the 70 MeV protons.