{"id":"22c47727-79f7-4082-8322-903597114f19","arxiv_id":"2411.17198","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In a ground-level-like mixed radiation field, 65 nm 6T SRAM cells show about 1.45x higher SEU cross-section than 8T cells, supporting the use of these memories as radiation monitors.","lead":"Researchers measured how often two types of memory cells, 6T and 8T SRAM, flip when exposed to a radiation field similar to ground-level cosmic rays. They found that the older 6T design is about 1.45 times more sensitive to single-event upsets than the newer 8T design, despite the 8T cell being larger.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The absolute per-bit cross-sections depend on a FLUKA-derived HEHeq fluence with no stated systematic uncertainty; a conversion-factor error would shift all values and the ground-level comparison, although the 6T/8T ratio is largely immune.","rationale":"The reader's weakest_assumption correctly identifies the FLUKA-derived HEHeq fluence as the load-bearing premise for the absolute cross-section values and for the 'ground level' label. My review reaches the same conclusion: the measured quantity is event counts, and converting those counts into a cross-section requires a fluence that is simulated, not locally measured. The paper gives no systematic uncertainty for this step, so the absolute numbers have an unquantified systematic error that could be as large as tens of percent. This does not invalidate the 6T versus 8T ratio, since both arrays are exposed to the same fluence and the ratio is insensitive to a common multiplicative factor. It does, however, affect the absolute cross-sections, the comparison with the literature in Table IV, and the proposed monitor use. I also note that the second-run data contain an internal inconsistency: the text reports 105 errors in 8T cells while Table III lists 70 events and 71 affected bits for SRAM8T; this further weakens the validation run but is secondary to the fluence issue. Because the reader already returned a CONDITIONAL verdict and my concern supports that assessment rather than overturning it, the appropriate verdict remains CONDITIONAL, so verdict_should_be is UNCHANGED.","tokens_in":12459,"tokens_out":4690,"duration_ms":45797,"concrete_test":"Perform a third irradiation run at the same CHARM position with an independently calibrated fluence monitor—for example, an LHC RadMon SRAM with known HEH cross-section or activation foils—placed immediately adjacent to the DUT. Derive the HEHeq fluence from this independent monitor and recompute the 6T and 8T cross-sections. If the FLUKA-based fluence differs from the independent measurement by more than the reported Poisson error (roughly 8%), the absolute cross-sections in Section 4 need to be revised and the ground-level comparison should be re-evaluated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim in Section 4—sigma_SEU = 7.5e-13 cm2 for 6T and 5.2e-13 cm2 for 8T under a ground-level-like mixed field—is computed via Eq. (8) using the HEHeq fluence. That fluence is not measured at the DUT; it is obtained by multiplying the CHARM proton beam intensity by a conversion factor from FLUKA simulations, as stated in Section 4. The paper reports only the 7–8% Poisson counting uncertainty and gives no systematic uncertainty for this conversion factor or for the spectral-weighting correction in Eq. (6). FLUKA predictions for CHARM test positions involve target, shielding, and geometry modeling, and the intermediate-neutron correction is known to be technology- and spectrum-dependent. A 20–40% error in the conversion factor would proportionally change both cross-sections and the claimed agreement with Table IV, and would weaken the assertion that the measured values represent the ground-level environment. The 6T/8T ratio (1.45x) is much more robust because it divides out the common fluence, so the relative finding is not the main point of vulnerability; the absolute cross-sections and the proposed use as HEH monitors are.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports accelerated mixed-field irradiation experiments performed at CERN's CHARM facility to characterize the SEU response of minimum-sized 6T and 8T SRAM bit-cells fabricated in a 65 nm CMOS technology. From a first 60-hour run, the authors derive per-bit SEU cross-sections of 7.5e-13 cm2 for 6T cells and 5.2e-13 cm2 for 8T cells, giving a 1.45x higher sensitivity for 6T cells, and they report a second 12-hour run with a similar ratio. The paper also analyzes multiple-cell upsets and proposes that the tested SRAMs could serve as high-energy-hadron (HEH) fluence monitors in HEP environments.","tokens_in":12848,"tokens_out":5016,"duration_ms":42825,"significance":"If the absolute cross-sections are reliable, the data provide a useful experimental point for ground-level soft-error rate estimation and for SRAM-based radiation monitoring in accelerator environments. The 6T/8T comparison is the paper's strongest contribution because the common-mode FLUKA-derived fluence cancels in the ratio, and the first-run event counts (214 vs. 147) give reasonable statistical significance. The main weakness is that the absolute cross-sections and the proposed ground-level equivalence rest on a FLUKA-derived HEHeq fluence for which no systematic uncertainty is reported.","major_comments":[{"comment":"The reported cross-sections are computed from N_SEU = sigma * Phi_HEHeq, where the HEHeq fluence is obtained by multiplying the CHARM proton beam intensity by a conversion factor from FLUKA simulations, as described in Section 4. The paper quotes only the Poisson counting uncertainty (7-8%) and gives no systematic uncertainty for the conversion factor or for the spectral-weighting correction in Eq. (6). Since any error in this factor scales all absolute cross-sections linearly, the authors should provide an uncertainty estimate or quantitative bound for the fluence and propagate it into the cross-section values in Section 4 and Table IV.","section":"Section 4 (fluence estimation and Eq. (8))"},{"comment":"The text states that 105 errors were detected in 8T cells in the second run, but Table III lists 70 total events for SRAM8T. The quoted 8T cross-section of 5.6e-13 cm2 corresponds to 70 events, not 105. This internal inconsistency must be corrected; if 105 events were actually counted, the 8T cross-section would be near the 6T value, contradicting the stated 1.5x ratio.","section":"Section 4 (second irradiation period and Table III)"},{"comment":"The final cross-section values are not reconciled between the two irradiation runs and Table IV. The first run gives 7.5e-13 cm2 (6T) and 5.2e-13 cm2 (8T), the second run gives 8.5e-13 cm2 and 5.6e-13 cm2, and Table IV reports 7.6e-13 cm2 and 5.2e-13 cm2 without stating which run or averaging method was used. The paper should explicitly state how the reported values were obtained and quantify the run-to-run spread as part of the uncertainty.","section":"Section 4 and Table IV"}],"minor_comments":[{"comment":"There are small language and typographical issues: 'iso-thermal and thermal part' should presumably be 'epithermal and thermal part', and 'cooper target' should be 'copper target'.","section":"Section 2 and 3.2"},{"comment":"The phrase 'as close as possible to the atmospheric environment' should be accompanied by a caveat that the CHARM mixed field is an accelerator-based representation, not a true ground-level exposure, to avoid overstating the ground-level label.","section":"Section 4"},{"comment":"The sentence 'The relative uncertainty of the measurements is given by 1/sqrt(N)' would be clearer if it stated the confidence level or standard deviation implied by this expression and explicitly accounted for the separate uncertainties in the two cross-section values.","section":"Section 4"},{"comment":"The text describing the second run says '105 errors in 6T cells' and '105 errors in 8T cells'; using 'errors' and 'events' interchangeably can be confusing because Table III distinguishes total events from affected bits. Please align the terminology.","section":"Section 4"},{"comment":"The comparison with literature values in Table IV is qualitative because the referenced experiments use different particle types, energies, and technology nodes; the statement that the results are 'in agreement' should be accompanied by this caveat.","section":"Table IV"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and the experimental effort is credible, but the missing systematic uncertainty on the FLUKA-derived fluence and the internal inconsistency in the second-run 8T count are load-bearing for the absolute cross-section claims. The 6T/8T ratio itself is more robust and could stand with a clearer uncertainty treatment. No concerns about novelty or citation behavior beyond the heavy reliance on the authors' own prior characterizations, which is appropriate here."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Main take: the paper gives the first comparison of 6T and 8T SRAM SEU cross-sections in the CHARM mixed field, and the ratio result (6T about 1.45x more sensitive) is a solid experimental finding. The absolute cross-sections carry an unquantified systematic from the FLUKA-based fluence conversion, and there is a clear text/table inconsistency in the second run. Both are fixable.\n\nWhat is actually new: prior work by the same group used mono-energetic neutron and proton beams; this is the first mixed-field dataset intended to represent a ground-level-like spectrum. The measurements are direct: N divided by fluence times bit count, no fitted parameters. The MCU analysis is low-statistics but supports the ratio trend. The guard-drain explanation for 8T robustness is plausible.\n\nSoft spots: (1) The HEHeq fluence at the DUT is not measured but derived from proton beam current times a FLUKA conversion factor, and no systematic uncertainty is reported. A 20–40% error in that factor would shift both cross-sections and the claimed agreement with Table IV. The 6T/8T ratio is unaffected, so the main qualitative conclusion holds, but the absolute values should be labeled as facility-limited until an uncertainty estimate is provided. (2) In the second run, the text says 105 errors in both 6T and 8T, but Table III lists 70 total events for 8T; the quoted cross-section for 8T is consistent with 69–70 events, so the \"105\" is likely a typo. The authors should correct it. (3) The comparison to prior work in Table IV is loose—absolute values from other groups span an order of magnitude, so \"in agreement\" should be argued more carefully, perhaps against their own mono-energetic data [34] rather than a wide reference spread.\n\nThe paper deserves a serious referee. The data are new and the ratio result is useful for designers and for radiation monitoring applications. With the inconsistency fixed and a stated systematic uncertainty, this would be a solid contribution to the radiation-effects literature.","headline":"New CHARM mixed-field data for 6T vs 8T SRAM; ratio robust, absolute values need systematic uncertainty.","tokens_in":13266,"tokens_out":2910,"would_cite":true,"duration_ms":25586,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A 65 nm SRAM comparison shows 6T cells have a 1.45x higher single-event upset cross-section than 8T cells in a ground-level-like mixed radiation field, despite nearly equal critical charge.","keywords":["Single Event Upset","SEU","SRAM","8T SRAM","mixed-field irradiation","soft-error cross-section","65 nm CMOS","radiation monitor"],"falsifier":"Irradiate the same 65 nm 6T and 8T arrays in a well-characterized mono-energetic neutron beam with fluence set by an independent dosimeter, measure the cross-section, and compare with the values implied by the mixed-field calibration. If the absolute values or the 1.45x ratio differ by more than the combined statistical errors (about 7-8% here), the simulated fluence conversion is wrong.","tokens_in":15,"feed_emoji":"☢️","tokens_out":7926,"duration_ms":130700,"temperature":0.7,"pith_summary":"This paper measures how often two SRAM memory cell designs, standard 6-transistor (6T) and 8-transistor (8T) cells, flip state under cosmic-ray-like radiation at ground level. In accelerated tests in a mixed hadron field built to mimic the atmospheric environment, the per-bit upset cross-section was 7.5e-13 cm2 for 6T cells and 5.2e-13 cm2 for 8T cells, a 1.45x advantage for 8T. Because the two cells have nearly identical critical charge, the result is evidence that critical charge alone does not predict radiation hardness; layout and parasitic charge-collection paths matter. The paper also argues the 6T memory is sensitive and stable enough to serve as a practical monitor of high-energy hadron fluence in accelerator environments.","feed_headline":"8T SRAM cells resist cosmic-ray upsets 1.45x better than 6T","feed_subtitle":"Per-bit cross-sections: 7.5e-13 cm2 for 6T, 5.2e-13 cm2 for 8T; 8T also yields fewer multi-bit events.","key_machinery":"The central object is the per-bit soft-error cross-section, $\\sigma_{SEU} = N_{SE}/\\Phi_{HEHeq}$, i.e., counted upsets divided by the equivalent high-energy-hadron fluence. The fluence is obtained by monitoring the primary proton beam and converting to HEHeq through Monte Carlo simulation of the test position's mixed field, including the intermediate-energy-neutron contribution. On the device side, the distinguishing feature is the 8T cell's two extra read-port transistors: during irradiation they are off, so they do not hold state but can collect charge, functioning as parasitic guard drains that reduce the probability of flipping the storage nodes.","core_discovery":"Under irradiation by a mixed hadron field whose energy spectrum is chosen to resemble the sea-level cosmic-ray environment, the measured per-bit single-event upset cross-sections are 7.5e-13 cm2 for minimum-size 6T cells and 5.2e-13 cm2 for 8T cells, a ratio of 1.45. Multi-cell upsets are also more frequent in 6T cells (4.3% vs 3.5% of events in the first run; 7.6% vs 1.4% in the second), and a repeat run confirmed the 1.5x ratio of cross-sections. The two cell types have practically the same critical charge (1.96 fC vs 2.07 fC), so the paper concludes that the difference comes from the read-port transistors of the 8T cell, which float in cut-off during irradiation and act as guard drains that collect part of the deposited charge.","pith_inferences":["A direct test of the guard-drain explanation would be to add grounded dummy nMOS drains to the 6T layout (mimicking the 8T read port) and check whether its cross-section falls to the 8T level; the paper's mechanism predicts it will.","The absolute cross-section numbers may be revised if the facility's simulated field-to-fluence calibration is updated, so inter-facility comparisons should quote the same HEHeq normalization; a mono-energetic calibration campaign would anchor the scale.","Because critical charge drops with supply voltage, the 6T/8T ratio may change at lower Vdd; testing at 0.9 V or 0.8 V would show whether the guard-drain advantage grows, shrinks, or inverts.","If the 8T multi-cell-upset fraction advantage holds at higher statistics, 8T memories would reduce the burden on multi-bit error-correction and scrubbing in radiation-monitoring and high-energy physics readout systems."],"forward_implications":["If the measured cross-sections are right, a 16 Mbit 6T memory of this design would suffer roughly 2.5 soft errors per year at sea level, a low rate for most consumer chips but measurable enough for monitoring.","8T cells, at 1.39x the layout area, buy a 1.45x lower SEU cross-section and a lower multiple-cell-upset fraction, so memory designs that can pay the area get both stability and soft-error benefit.","Because critical charge is nearly equal (1.96 vs 2.07 fC) yet cross-sections differ by 1.45x, the study reinforces that radiation hardness must be evaluated experimentally; critical-charge-only ranking will misrank cells.","The 6T SRAM's combination of thermal-neutron insensitivity and fluence resolution around 8e4 cm2/count makes it a candidate HEH fluence monitor in accelerator tunnels and high-energy physics service areas.","The observed multi-cell upsets all occupied adjacent physical locations, so error-correction codes that correct nearby physical neighbors will be sufficient for these cells."],"supporting_citations":[{"why":"Supplies the simulated hadron energy spectra at the test position and the comparison establishing that the spectrum resembles the atmospheric environment.","marker":"[18]"},{"why":"Provides the Monte Carlo transport calculation used to convert monitored proton-beam intensity into the HEHeq fluence that normalizes every cross-section.","marker":"[29]"},{"why":"Earlier irradiation of the same 6T/8T designs with mono-energetic neutrons and protons; gives the 1.5x ratio reference and evidence of thermal-neutron insensitivity.","marker":"[34]"},{"why":"Prior alpha-particle study finding 6T more sensitive than 8T and showing critical charge does not rank SER, grounding the paper's explanation of the ratio.","marker":"[11]"},{"why":"Quantifies how off-state transistors can act as guard drains that collect charge, the mechanism invoked to explain the 8T cell's lower cross-section.","marker":"[32]"},{"why":"Describes the existing SRAM-based radiation monitoring system whose fluence-resolution requirements the tested 6T memory is compared against.","marker":"[12]"},{"why":"Characterizes the mixed-field facility and its representative test positions, supporting the claim that the exposure approximates ground-level cosmic rays.","marker":"[28]"},{"why":"Establishes the Poisson-event assumption used to assign the 7-8% statistical counting uncertainty to the measured cross-sections.","marker":"[30]"}],"fun_headline_variants":["8T SRAM beats 6T by 1.45x in cosmic-ray tests","8T SRAM cells show 1.45x lower SEU cross-section","Guard drains in 8T SRAM lower SEU cross-section by 1.45x","8T SRAM cells: 1.45x fewer upsets, ideal for radiation monitors"],"cache_read_input_tokens":15488,"weakest_assumption_plain":"All the cross-section numbers are scaled by a simulation-calculated factor that converts accelerator beam intensity into the hadron fluence the chip actually saw; if that factor is wrong, every per-bit cross-section in the paper is wrong by the same factor.","fun_headline_variants_meta":{"raw":{"variants":["8T SRAM beats 6T by 1.45x in cosmic-ray tests","8T SRAM cells show 1.45x lower SEU cross-section","Guard drains in 8T SRAM lower SEU cross-section by 1.45x","8T SRAM cells: 1.45x fewer upsets, ideal for radiation monitors"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001154,"raw_usage":{"total_tokens":4771,"prompt_tokens":922,"completion_tokens":3849,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":538,"completion_tokens_details":{"reasoning_tokens":3754}},"tokens_in":538,"tokens_out":3849,"duration_ms":24991,"temperature":1.0,"reasoning_tokens":3754,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T12:23:15.702686+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Irradiate the same 65 nm 6T and 8T arrays in a well-characterized mono-energetic neutron beam with fluence set by an independent dosimeter, measure the cross-section, and compare with the values implied by the mixed-field calibration. If the absolute values or the 1.45x ratio differ by more than the combined statistical errors (about 7-8% here), the simulated fluence conversion is wrong.","supporting_citations":[{"cited_title":"Monte Carlo Evaluation of Single Event Effects in a Deep - Submicron Bulk Technology: Comparison Between Atmospheric a nd Accelerator Environment,","cited_arxiv_id":null,"evidence_quote":"Supplies the simulated hadron energy spectra at the test position and the comparison establishing that the spectrum resembles the atmospheric environment."},{"cited_title":"FLUKA Simulations for SE E Studies of Critical LHC Underground Areas,","cited_arxiv_id":null,"evidence_quote":"Provides the Monte Carlo transport calculation used to convert monitored proton-beam intensity into the HEHeq fluence that normalizes every cross-section."},{"cited_title":"Soft error rate comparison of 6T and 8T SRAM ICs using mono -energetic proton and neutron irradiation sources","cited_arxiv_id":null,"evidence_quote":"Earlier irradiation of the same 6T/8T designs with mono-energetic neutrons and protons; gives the 1.5x ratio reference and evidence of thermal-neutron insensitivity."},{"cited_title":"Detailed 8-transistor SRAM cell analysis for improved alpha particle radiation hardening in nanometer technologies","cited_arxiv_id":null,"evidence_quote":"Prior alpha-particle study finding 6T more sensitive than 8T and showing critical charge does not rank SER, grounding the paper's explanation of the ratio."},{"cited_title":"Failure map functions and accelerated mean time to failure tests: New approaches from improving. the reliability stimation in systems exposed to single event upsets,","cited_arxiv_id":null,"evidence_quote":"Establishes the Poisson-event assumption used to assign the 7-8% statistical counting uncertainty to the measured cross-sections."}],"review_version":1}