40 CURATED RECORDS · EVIDENCE LEDGER筆收錄紀錄 · 證據總帳

不是來源清單
是證據邊界
Not a reading list
An evidence boundary

每一筆資料都回答三件事:它直接證明什麼、不能證明什麼,以及它在 Secure Storage 論述中應該站在哪一層。Every record states what it directly supports, what it does not establish, and where it belongs in the Secure Storage argument.

12ATTACK / CIRCUIT PAPERS攻擊/電路論文
5PATENT MECHANISMS專利機制
22VENDOR / OFFICIAL SOURCES供應商/官方來源
1OFFICIAL CASE官方案例

EVIDENCE CLASS · NOT ASSURANCE MATURITY證據類型 · 非保證成熟度

主張 → 證據 → 推論,
不跨級偷換
Claim → Evidence → Inference,
without category errors

這裡分類「來源如何支持主張」,不同於 Assurance Center 的 M1–M6 成熟度。專利用於證明 sensing topology 的存在;供應商文件描述自家 architecture 與測試主張;攻擊難度與普遍性只由獨立實證支撐。This classifies how a source supports a claim; it is distinct from the Assurance Center's M1–M6 maturity. Patents establish that a sensing topology exists; vendor documents describe their architecture and test claims; only independent demonstrations support attack feasibility and generality.

E1 · DIRECT OBSERVATIONE1 · 直接觀測

實體攻擊示範Physical demonstration

有目標晶片、設備、程序與恢復結果。Target chip, equipment, procedure and recovered result.

E2 · MECHANISM EVIDENCEE2 · 機制證據

電路與物理Circuit and physics

說明 read、leakage、breakdown 或 sensing 如何發生。Explains how read, leakage, breakdown or sensing occurs.

E3 · VENDOR DISCLOSUREE3 · 供應商揭露

產品公開主張Product disclosure

可引用,但需標記為供應商自己提供的證據。Citable with explicit attribution as vendor-provided evidence.

E4 · BOUNDED INFERENCEE4 · 有界推論

系統級結論System inference

由多筆來源組成,仍需產品或節點實測封閉。Composed from sources and closed only by product/node validation.

ANNOTATED SOURCE CORPUS附註來源彙編

查證每一句話,
也查證它的限制
Verify every sentence—
and its limit

使用搜尋或類型篩選。關鍵詞同時涵蓋技術、攻擊、作者、DOI、PUF、感測放大器、光子放射、FIB 與產品架構。Search or filter by source type. Keywords cover technology, attack, author, DOI, PUF, sense amplifiers, photon emission, FIB and product architecture.

日期閱讀方式:來源發表日期與版本見各筆書目;CHECKED 是該筆原有查核日期,不等同本頁修改日,也不代表每項主張都已逐一確證。未標 CHECKED 者保留原初查閱脈絡,不能推定已重新查核。2026-10-08 只記錄本次文字與證據界線修訂,未新增外部來源查證。Date contract: publication date and version belong to each source record. CHECKED retains that record's review date; it is neither the page modification date nor proof that every claim was individually established. Records without CHECKED retain the original review context, without implying a fresh check. 2026-10-08 records wording and evidence-boundary changes only; no new external-source verification was performed.

40 筆符合目前條件records match the current viewINITIAL REVIEW 2026-08-12 · RECORD CHECKS THROUGH 2026-10-06 · PAGE MODIFIED 2026-10-08初始查閱 2026-08-12 · 個別查核截至 2026-10-06 · 頁面修改 2026-10-08
P01PEER-REVIEWED同儕審查

2009 · IEEE TRANSACTIONS ON COMPUTERS · E2

Power-Up SRAM State as an Identifying Fingerprint and Source of True Random Numbers

Holcomb、Burleson 與 Fu 建立 SRAM power-up fingerprint 的實驗基礎,將 device identity 連結到 MOSFET threshold mismatch 與 startup noise。Holcomb, Burleson and Fu establish the experimental basis for SRAM power-up fingerprints rooted in MOSFET threshold mismatch and startup noise.

支持 (SUPPORTS)SUPPORTS

SRAM PUF 的物理來源與 power-up expression。Physical origin and power-up expression of SRAM PUF.

不證明 (DOES NOT PROVE)DOES NOT PROVE

特定商用 PUF 的完整攻擊抵抗力。Full attack resistance of a commercial PUF.

DOI 10.1109/TC.2008.212 ↗
P02PEER-REVIEWED同儕審查

2011 · IEEE TVLSI · E2

Design and Sensitivity Analysis of a New Current-Mode Sense Amplifier for Low-Power SRAM

提出 SRAM current-mode sense amplifier,直接反證「SRAM 本質上只能 voltage sensing」。Presents a current-mode SRAM sense amplifier, directly disproving that SRAM is inherently voltage-sense only.

支持 (SUPPORTS)SUPPORTS

Sense mode 是 circuit choice。Sense mode is a circuit choice.

不證明 (DOES NOT PROVE)DOES NOT PROVE

任何特定 SRAM macro 採用此 topology。That a specific SRAM macro uses it.

DOI 10.1109/TVLSI.2009.2033110 ↗
T01PATENT專利

2009 · US7511982B2 · E2

High Speed OTP Sensing Scheme

Sidense 專利揭露 twin-cell differential arrangement:一對 cell 中恰有一個被 programming,以 complementary bitline 的 voltage difference 進行正常讀取。Sidense discloses a twin-cell differential arrangement with exactly one programmed cell per pair and normal read through complementary-bitline voltage difference.

支持 (SUPPORTS)SUPPORTS

1T twin-cell、真正互補 representation 與 differential read 的技術基礎。The technical basis for 1T twin-cell complementary representation and differential read.

不證明 (DOES NOT PROVE)DOES NOT PROVE

所有 Synopsys macro 的預設設定或精確 macro area。The default configuration of every Synopsys macro or exact macro area.

Google Patents · US7511982B2 ↗
T02PATENT專利

1993 · US5247479A · E2

Current Sensing Amplifier for SRAM

Intel 專利將 SRAM bitline voltage differential 轉為 current differential,再轉回放大的 voltage output。Intel's patent converts SRAM bitline voltage differential to current differential and then back to an amplified voltage output.

支持 (SUPPORTS)SUPPORTS

Voltage/current 是連續 read chain 的不同節點。Voltage/current are nodes on one read chain.

不證明 (DOES NOT PROVE)DOES NOT PROVE

current-mode 比 voltage-mode 更不安全或更安全。That current mode is more or less secure.

Google Patents · US5247479A ↗
P03ATTACK STUDY攻擊研究

2018 · ISTFA · E1

Comparative Assessment of Optical Techniques for Semi-Invasive SRAM Data Read-out

在商用 MSP430 比較 Photon Emission、TLS 與 EOFM,說明 SRAM content readout 對 core voltage 與 modulation 條件的依賴。Compares Photon Emission, TLS and EOFM on a commercial MSP430, showing SRAM readout dependence on core voltage and modulation.

支持 (SUPPORTS)SUPPORTS

三類 optical failure-analysis technique 可讀 SRAM。Three optical FA techniques can read SRAM.

不證明 (DOES NOT PROVE)DOES NOT PROVE

所有節點與封裝都有相同 SNR。Equal SNR across nodes and packages.

DOI 10.31399/asm.cp.istfa2018p0266 ↗
P04ATTACK STUDY攻擊研究

2018 · IEEE HOST · E1

Direct Read of Idle Block RAM from FPGAs Utilizing Photon Emission Microscopy

以 SRAM cell 內 state-dependent NMOS leakage photon 直接讀取 idle FPGA block RAM,不需啟動正常 read circuitry。Reads idle FPGA block RAM through state-dependent NMOS leakage photons without activating normal read circuitry.

支持 (SUPPORTS)SUPPORTS

SRAM 靜態 state 也可能形成位置可辨光子訊號。Static SRAM state can create spatially distinguishable photons.

不證明 (DOES NOT PROVE)DOES NOT PROVE

斷電且無 remanence 時仍能讀 0/1。0/1 recovery after power-off without remanence.

DOI 10.1109/HST.2018.8383889 ↗
P05ATTACK STUDY攻擊研究

2021 · CRYPTOGRAPHY AND COMMUNICATIONS · E1

Quiescent Photonics Side Channel Analysis: Low-Cost SRAM Readout Attack

在 PIC 與 ARM Cortex-M0 展示 quiescent photon differential analysis,並提出較低成本的 sample preparation 與 imaging setup。Demonstrates quiescent photon differential analysis on PIC and ARM Cortex-M0 devices with lower-cost preparation and imaging.

支持 (SUPPORTS)SUPPORTS

光子側信道不只存在於高階實驗室。Photon side channels are not limited to top-tier labs.

不證明 (DOES NOT PROVE)DOES NOT PROVE

對所有 advanced node 都低成本。Low cost on every advanced node.

DOI 10.1007/s12095-020-00469-5 ↗
P06ATTACK STUDY攻擊研究

2018 · IACR TCHES · E1

Key Extraction Using Thermal Laser Stimulation: Xilinx UltraScale FPGAs

從 battery-backed SRAM 以一次非侵入量測恢復 256-bit AES key,並討論 power-off TLS countermeasure。Recovers a 256-bit AES key from battery-backed SRAM in one non-invasive measurement and presents a power-off TLS countermeasure.

支持 (SUPPORTS)SUPPORTS

把 key 放在 powered SRAM 並不免疫 failure-analysis tools。Powered SRAM key storage is not immune to FA tools.

不證明 (DOES NOT PROVE)DOES NOT PROVE

等同攻破 SRAM PUF reconstruction。Equivalence to breaking SRAM-PUF reconstruction.

IACR ePrint 2018/717 ↗
P07ATTACK STUDY攻擊研究

2002 · UNIVERSITY OF CAMBRIDGE · E1

Low Temperature Data Remanence in Static RAM

實驗建立現代 SRAM retention time 與低溫的關係,挑戰「斷電即消失」的簡化假設。Experimentally relates SRAM retention time to low temperature, challenging the simplified “power-off means immediate loss” assumption.

支持 (SUPPORTS)SUPPORTS

Volatile 不等於零 remanence window。Volatile does not mean zero remanence window.

不證明 (DOES NOT PROVE)DOES NOT PROVE

正常環境下 SRAM 長期保留。Long-term SRAM retention at normal conditions.

DOI 10.48456/tr-536 ↗
P08ATTACK STUDY攻擊研究

2019 · MICROPROCESSORS AND MICROSYSTEMS · E1

Attacking SRAM PUFs Using Very-Low-Temperature Data Remanence

研究 −110°C 至 −40°C 的 intrinsic SRAM PUF remanence,顯示攻擊者可高機率得知 response,並討論 countermeasure。Studies intrinsic SRAM-PUF remanence from −110°C to −40°C, showing high-probability response recovery and discussing countermeasures.

支持 (SUPPORTS)SUPPORTS

SRAM PUF 的「absent」需要正常條件邊界。“Absent” needs a normal-conditions boundary.

不證明 (DOES NOT PROVE)DOES NOT PROVE

商用 IP 未採取偵測或抵抗措施。That commercial IP lacks detection or resistance.

DOI 10.1016/j.micpro.2019.102864 ↗
P09ATTACK STUDY攻擊研究

2025 · USENIX WOOT · E1

Extraction of Secrets from 40 nm CMOS Gate-Dielectric-Breakdown Antifuses by FIB PVC

在 RP2350 類 40 nm commodity MCU 示範 FIB passive voltage contrast 擷取 antifuse OTP,並量化 sample preparation 與分析時間。Demonstrates FIB passive-voltage-contrast extraction of antifuse OTP on a 40 nm commodity MCU and reports preparation/analysis effort.

支持 (SUPPORTS)SUPPORTS

Antifuse 不應被視為不可讀的明文 key vault。Antifuse should not be treated as an unreadable plaintext key vault.

不證明 (DOES NOT PROVE)DOES NOT PROVE

所有 antifuse macro 成本相同或都能完整逐 bit 恢復。Equal cost or complete bit recovery for every antifuse macro.

USENIX WOOT 2025 ↗
P10ATTACK STUDY攻擊研究

CAMBRIDGE REPOSITORY · E1

Reverse Engineering Flash EEPROM Memories Using Scanning Electron Microscopy

backside preparation 後以 SEM passive voltage contrast 區分 floating-gate cell 0/1,並示範自動化 full-memory acquisition。Uses SEM passive voltage contrast after backside preparation to distinguish floating-gate 0/1 and automate full-memory acquisition.

支持 (SUPPORTS)SUPPORTS

Persistent charge 也能成為 failure-analysis contrast。Persistent charge can become FA contrast.

不證明 (DOES NOT PROVE)DOES NOT PROVE

與 antifuse、eFuse 攻擊程序完全相同。Identical procedure for antifuse or eFuse.

University of Cambridge Repository ↗
P11PEER-REVIEWED同儕審查

2021 · APPLIED SCIENCES · E2

Balancing the Leakage Currents in Nanometer CMOS Logic—A Challenging Goal

分析 nanometer CMOS data-dependent static leakage 與 countermeasure,支持「靜態 CMOS 也有 current side channel」。Analyzes data-dependent static leakage in nanometer CMOS and countermeasures, supporting that static CMOS also has a current side channel.

支持 (SUPPORTS)SUPPORTS

靜態 power signature 與邏輯狀態相關。Static power signatures can depend on logic state.

不證明 (DOES NOT PROVE)DOES NOT PROVE

特定 memory array 可直接影像化。Direct imaging of a specific memory array.

DOI 10.3390/app11157143 ↗
T03PATENT專利

2022 · US20220366113A1 · E2

Protecting Against Emission-Based Side-Channel Detection

以 SRAM static photon emission、EOFM/LVI 等威脅為出發點提出 countermeasure,說明 industry 將此視為實際設計問題。Proposes countermeasures against SRAM static photon emission and EOFM/LVI, showing that industry treats it as a design concern.

支持 (SUPPORTS)SUPPORTS

SRAM emission attack 有專利級防禦需求。SRAM emission attacks motivate patent-level defenses.

不證明 (DOES NOT PROVE)DOES NOT PROVE

某產品已實作該 countermeasure。That a product implements the countermeasure.

Google Patents · US20220366113A1 ↗
T04PATENT專利

2015 · US20150234751A1 · E2

System for Generating a Cryptographic Key from a Memory Used as a PUF

討論 photonic-emission 對 SRAM PUF first-read/power-up content 的潛在攻擊與防護概念。Discusses potential photonic-emission attacks on first-read/power-up SRAM-PUF contents and protective concepts.

支持 (SUPPORTS)SUPPORTS

「key 不永久儲存」仍需保護 response capture。A non-stored key still requires response-capture protection.

不證明 (DOES NOT PROVE)DOES NOT PROVE

攻擊可繞過現代商用 PUF 全部 countermeasure。Bypass of all modern commercial-PUF countermeasures.

Google Patents · US20150234751A1 ↗
V01VENDOR BOOK供應商書籍

PUFSECURITY · E3

Quantum Tunneling PUF: Basics, Circuits, and Security Applications

描述 NeoPUF enrollment、永久 tunneling path、single-ended Icell/Iref sensing、SEM/TEM 與 InGaAs 測試,以及其對 SRAM PUF 的比較。Describes NeoPUF enrollment, persistent tunneling paths, single-ended Icell/Iref sensing, SEM/TEM and InGaAs tests, and comparison with SRAM PUF.

支持 (SUPPORTS)SUPPORTS

NeoPUF 自述機制與內部測試結果。NeoPUF's disclosed mechanism and internal tests.

不證明 (DOES NOT PROVE)DOES NOT PROVE

第三方獨立重現或所有節點免疫。Independent replication or immunity across nodes.

PUFsecurity PDF · pp. 46–97, 108–112 ↗
V02PRODUCT PAGE產品頁面

SYNOPSYS · E3

Secure Storage Solution for OTP IP

公開架構:antifuse OTP、SRAM PUF power-up root、AES-256 engine、address scrambling、Secure Controller、APB 與 automatic provisioning。Discloses antifuse OTP, SRAM-PUF power-up root, AES-256 engine, address scrambling, Secure Controller, APB and automatic provisioning.

支持 (SUPPORTS)SUPPORTS

產品 scope、interfaces 與公開 benefit。Product scope, interfaces and disclosed benefits.

不證明 (DOES NOT PROVE)DOES NOT PROVE

未公開 crypto mode、integrity 與 attack-test 細節。Undisclosed crypto mode, integrity and attack-test details.

Synopsys Product Page ↗
V03TECHNICAL ARTICLE技術文章

SYNOPSYS · E3

Strengthening Silicon-Level Security with the Secure Storage Solution for OTP IP

把 FIB/physical access 與 antifuse plaintext risk 轉成 encrypted OTP + non-stored root 的 layered-defense 主張。Frames FIB/physical access and antifuse plaintext risk as a layered-defense case for encrypted OTP and a non-stored root.

支持 (SUPPORTS)SUPPORTS

產品設計意圖與威脅敘事。Product design intent and threat narrative.

不證明 (DOES NOT PROVE)DOES NOT PROVE

完整 security proof 或特定 attacker cost。A complete security proof or attacker cost.

Synopsys Technical Article ↗
V04TECHNICAL ARTICLE技術文章

SYNOPSYS · E3

Understanding SRAM PUF: The Secure Silicon Fingerprint

說明 uninitialized SRAM、Activation Code 儲存位置、key on-demand extraction,以及斷電時 key 不存在的產品模型。Explains uninitialized SRAM, Activation Code storage, on-demand key extraction and the product model in which the key is absent at power-off.

支持 (SUPPORTS)SUPPORTS

Synopsys SRAM PUF lifecycle 與 integration requirements。Synopsys SRAM-PUF lifecycle and integration requirements.

不證明 (DOES NOT PROVE)DOES NOT PROVE

Activation Code 的完整 leakage/integrity proof。Complete leakage/integrity proof for the Activation Code.

Synopsys SRAM PUF Article ↗
V05DATASHEET規格書

2024 · PUFSECURITY · E3

Secure OTP — Tamperproof Storage Datasheet

公開 1024-bit PUF、physical address scrambling、I/O shuffling、APB、permission/zeroization/lock 與 instant hardware encryption 等產品主張。Discloses a 1024-bit PUF, physical address scrambling, I/O shuffling, APB, permission/zeroization/lock and instant hardware-encryption claims.

支持 (SUPPORTS)SUPPORTS

Secure OTP 公開功能範圍。Public Secure OTP feature scope.

不證明 (DOES NOT PROVE)DOES NOT PROVE

crypto algorithm/mode 與完整 independent attack evidence。Crypto algorithm/mode or independent attack evidence.

PUFsecurity Secure OTP Datasheet ↗
P12FOUNDRY PAPER晶圓代工研究論文

2021 · TSMC RESEARCH · E2

A 16Kb 5nm Antifuse One-Time Programmable Memory with Pseudo-Differential Sensing

TSMC 公開 5 nm antifuse OTP macro,採 bootstrap high voltage、read endpoint detection 與 pseudo-differential sensing,並報告高溫十年 retention 目標。TSMC discloses a 5 nm antifuse OTP macro using bootstrap high voltage, read-endpoint detection and pseudo-differential sensing with a ten-year high-temperature retention target.

支持 (SUPPORTS)SUPPORTS

先進節點 antifuse 與 pseudo-differential sensing 的實作可行性。Feasibility of advanced-node antifuse and pseudo-differential sensing.

不證明 (DOES NOT PROVE)DOES NOT PROVE

商用 macro 的預設安全模式或特定 IP 授權合約。A commercial macro's default security mode or a specific IP license contract.

TSMC Research ↗
T05PATENT專利

2023 · US11735266B2 · E2

Antifuse-Type One-Time Programmable Memory Cell and Array

eMemory 專利的傳統 cell 圖示包含 select transistor 與 antifuse transistor/capacitor device,提供「select device + programming device」成本模型的一手依據。eMemory's conventional cell drawing includes a select transistor and an antifuse transistor/capacitor device, providing primary support for the select-device-plus-programming-device cost model.

支持 (SUPPORTS)SUPPORTS

為何真正 complementary pair 會增加 cell 與 routing burden。Why a true complementary pair adds cell and routing burden.

不證明 (DOES NOT PROVE)DOES NOT PROVE

每一代 NeoFuse 都採完全相同 topology,或可用 transistor ratio 推算 macro area。That every NeoFuse generation uses the same topology or that transistor ratio predicts macro area.

Google Patents · US11735266B2 ↗
V06OFFICIAL NEWS官方消息

2017-10-17 · SYNOPSYS · E3

Synopsys Acquires Sidense Corporation

Synopsys 官方收購公告明列 Sidense patented split-channel 1T-Fuse、低功耗/小面積、最快 10 ns read 與 volume-production 紀錄。Synopsys' acquisition release explicitly cites Sidense's patented split-channel 1T-Fuse, low power/small area, reads as fast as 10 ns and volume-production history.

支持 (SUPPORTS)SUPPORTS

1T-Fuse 的官方技術血統與 2017 收購年份。Official 1T-Fuse lineage and the 2017 acquisition date.

不證明 (DOES NOT PROVE)DOES NOT PROVE

每個後續 macro 都沿用相同 read configuration。That every later macro uses the same read configuration.

Synopsys Official Release ↗
V07TECHNICAL ARTICLE技術文章

SYNOPSYS · E3

Ensuring Reliable, Secure OTP IP in Advanced Nodes

Synopsys 公開說明其 OTP「可配置」為 differential read mode,以提高由 power signature 辨識內容的難度。Synopsys publicly states that its OTP can be configured in differential read mode to make content identification from power signature more difficult.

支持 (SUPPORTS)SUPPORTS

Differential read 是現行商業 feature 與安全選項。Differential read is a current commercial feature and security option.

不證明 (DOES NOT PROVE)DOES NOT PROVE

所有 macro 預設開啟、完全消除 optical/FIB leakage,或精確 PPA impact。Default enablement on every macro, elimination of optical/FIB leakage, or exact PPA impact.

Synopsys Technical Article ↗
V08OFFICIAL NEWS官方消息

2018-01-10 · SYNOPSYS · E3

Synopsys Acquires Kilopass Technology

官方公告確認 Kilopass 收購日期為 2018-01-10,並列出 1T/2T bitcell portfolio、10/7 nm silicon-proven 與超過 100 億顆出貨紀錄。The official release dates the Kilopass acquisition to 2018-01-10 and cites a 1T/2T portfolio, silicon-proven 10/7 nm offerings and more than 10 billion units shipped.

支持 (SUPPORTS)SUPPORTS

Kilopass advanced-node 能力、產品範圍與正確收購年份。Kilopass advanced-node capability, product scope and correct acquisition year.

不證明 (DOES NOT PROVE)DOES NOT PROVE

TSMC 特定 in-house macro 的授權條款或電路完全相同。Licensing terms or circuit identity for a specific TSMC in-house macro.

Synopsys Official Release ↗
V09FOUNDRY NEWS晶圓代工消息

2005 · TSMC · E3

Kilopass XPM Qualified on TSMC 0.18 µm Process

TSMC 官方新聞記錄 Kilopass XPM 在 TSMC 製程完成 qualification,建立雙方公開可追溯的早期合作關係。TSMC's official release records Kilopass XPM qualification on a TSMC process, establishing an early, publicly traceable relationship.

支持 (SUPPORTS)SUPPORTS

TSMC/Kilopass 的歷史合作與技術血統。Historical TSMC/Kilopass collaboration and technology lineage.

不證明 (DOES NOT PROVE)DOES NOT PROVE

10 nm 以下授權合約、延續條款或 5 nm macro 的直接來源。Sub-10 nm licensing terms, continuity clauses or the direct origin of the 5 nm macro.

TSMC Official Release ↗
V10供應商揭露VENDOR DISCLOSURE

2021-07-06 · eMEMORY & CHIPESTIMATE · E3

How eNVM Helps Power Controllers Be Smarter

eMemory 的原作者文章區分固定微調用 OTP 與可更新控制資料用 NeoMTP;NeoMTP 屬浮動閘 MTP。文中分別說明零額外光罩設計與採額外光罩縮小單元的後續版本,成本、耐久度及保存條件須綁定具名 IP、版本與製程驗證。 eMemory's author article distinguishes OTP for fixed trimming from NeoMTP for updateable control data; NeoMTP is floating-gate MTP. It describes a zero-mask design and a later version with additional masks for a smaller cell. Cost, endurance and retention require validation for the named IP, version and process.

支持 (SUPPORTS)SUPPORTS

OTP 與浮動閘 NeoMTP 的用途分流,以及具名 BCD 版本的製程整合與光罩取捨。Separate roles for OTP and floating-gate NeoMTP, plus process integration and mask tradeoffs for named BCD versions.

不證明 (DOES NOT PROVE)DOES NOT PROVE

所有 OTP 與 NeoMTP 版本共用相同物理機制、零光罩條件或高溫保存保證;PPA 與可靠度仍須依具名 IP 和晶圓廠 PDK 驗證。A common physical mechanism, zero-mask condition or high-temperature retention guarantee for every OTP and NeoMTP version; PPA and reliability still require named-IP and foundry-PDK validation.

ChipEstimate Technical Article ↗
V11INDUSTRY STANDARD產業標準

JEDEC STANDARDS & NAMED SUPPLIER PLAN · E3 / OFFICIALJEDEC 標準與具名供應商計畫 · E3/官方

DDR5 / HBM Post-Package Repair and the SK hynix–TSMC HBM4 Plan

DDR5、HBM3 與 HBM4 分別對應 JESD79-5、JESD238 與 JESD270-4 標準系列。hPPR 表示永久封裝後修復,其功能限制須核對標準版本與器件。另於 2024 年 4 月 19 日,SK hynix 與 TSMC 公告 HBM4 先進邏輯 Base Die/CoWoS 合作及 2026 年量產計畫;該公告不能證明計畫已完成。 DDR5, HBM3 and HBM4 correspond to the JESD79-5, JESD238 and JESD270-4 standard families. hPPR denotes permanent post-package repair; feature limits require the relevant revision and device. Separately, on April 19, 2024, SK hynix and TSMC announced HBM4 advanced-logic base-die / CoWoS collaboration with a 2026 mass-production plan. That announcement does not establish completion.

支持 (SUPPORTS)SUPPORTS

永久修復的功能概念,以及具名供應商於公告日期揭露的 HBM4 合作方向與時程計畫。The functional concept of permanent repair and the named suppliers’ HBM4 collaboration direction and schedule as disclosed on the announcement date.

不證明 (DOES NOT PROVE)DOES NOT PROVE

JEDEC 強制 AntiFuse、固定低於 100Ω 電阻、零回跳風險、所有 HBM4 供應商採相同節點或 0-mask OTP,以及探針卡採購全面移轉;亦不證明無限備援容量。A JEDEC mandate for AntiFuse, fixed resistance below 100Ω, zero grow-back risk, common nodes or 0-mask OTP across HBM4 suppliers, a universal probe-card procurement shift, or unlimited spare capacity.

JEDEC Official Specification ↗
SK hynix/TSMC 合作公告(2024-04-19)↗SK hynix / TSMC Announcement (2024-04-19) ↗
V12TECHNICAL DISCLOSURE技術揭露

SID & IEEE TRANSACTIONS · E2 / E3

High-Voltage Display Driver IC Optical Uniformity: De-Mura LUT & Gamma 2.2 Calibration

顯示技術學會 (SID) 與業界高壓 DDIC 論文揭露:AMOLED 螢幕因有機發光層厚度與驅動 TFT 閾值電壓離散,原始均勻度僅 70%~75%。透過在驅動 IC 內嵌 32Kb 至 128Kb MTP 儲存二維 Gain LUT,可將顯示均勻度拉升至 99% 以上,並將 Delta E 壓制在 0.5 以內。 SID and IEEE DDIC disclosures confirm that raw AMOLED panels exhibit 70%-75% optical uniformity due to TFT mobility and OLED layer variance. Embedding 32Kb-128Kb MTP within the driver IC to store 2D Gain LUTs elevates display uniformity beyond 99% and crushes Delta E below 0.5.

支持 (SUPPORTS)SUPPORTS

高壓 AMOLED 驅動晶片必須依賴高密度 eNVM 儲存光學補償查找表與 Gamma 校準碼。High-voltage AMOLED DDICs mandate dense eNVM for optical compensation LUTs and Gamma trims.

不證明 (DOES NOT PROVE)DOES NOT PROVE

靜態 LUT 能完全解決動態螢幕老化問題;動態老化仍需配合即時電流傳感閉環算法。Static LUT eliminates dynamic panel burn-in; real-time current sensing is required for aging.

IEEE Xplore Digital Library ↗
V13MARKET & TECHNICAL SURVEY市場與技術調查

2023 · E-INK ECOSYSTEM DISCLOSURE · E32023 · E-Ink 生態系揭露 · E3

E-Ink Ultra-HV Driving & MTP-to-OTP Product Lifecycle: 40V-50V Pulse & Waveform LUT

內部市場調查與領導驅動晶片廠商(天鈺 Fitipower、晶宏 UltraChip、晶門 Solomon)技術會議揭露:電子紙微膠囊電泳需 40V-50V(110HV/90HV)高壓脈衝,標準 BCD 因接面漏電過高無法採用。黑白邁向四色與全彩(Spectra 6/Kaleido 3)演進期採用 32Kb-64Kb MTP 儲存動態波形;成熟期為追求極致成本與 <250µm 窄高佈局全面轉入純邏輯 OTP。 Industry survey and design house disclosures (Fitipower, UltraChip, Solomon) confirm that electrophoretic displays require 40V-50V (110HV/90HV) pulses, precluding standard BCD due to excessive junction leakage. The color transition era mandates 32Kb-64Kb MTP for evolving waveforms, migrating to pure-logic OTP (<250µm macro height) in volume maturity.

支持 (SUPPORTS)SUPPORTS

電子紙驅動晶片排除一般 BCD 製程,且遵循「演進期 MTP ➔ 成熟期 OTP」產品戰略。E-Ink DDICs strictly exclude BCD and follow the 'MTP in transition ➔ OTP in maturity' roadmap.

不證明 (DOES NOT PROVE)DOES NOT PROVE

所有全彩彩色粒子配方均已停止疊代;全彩波形算法目前仍在動態演進中。All color particle formulations have ceased iteration; full-color waveforms remain evolving.

E Ink Official Technology ↗
V14供應商揭露VENDOR DISCLOSURE

2023-11-20 · SKYWATER / WEEBIT QUALIFICATION DISCLOSURE · E3 · CHECKED 2026-10-03

Weebit Nano / SkyWater: ReRAM Qualification for the S130 Process

SkyWater 與 Weebit 在 2023-11-20 的聯合公告指出:Weebit 的 1T1R ReRAM 巨集已在 SkyWater S130(130 nm)製程完成最高 125°C 的資格驗證,並納入該製程 PDK。此處記錄的是供應商對具名巨集與製程的揭露。The 20 November 2023 SkyWater / Weebit announcement reports qualification of the Weebit 1T1R ReRAM module in the SkyWater S130 (130 nm) process up to 125°C and its inclusion in that process PDK. This is a supplier disclosure for a named module and process.

支持 (SUPPORTS)SUPPORTS

具名 S130 ReRAM 巨集的資格驗證與 PDK 可用性聲稱。A supplier-reported qualification and PDK availability milestone for the named S130 ReRAM module.

不證明 (DOES NOT PROVE)DOES NOT PROVE

不能把此結果套用到其他代工廠、175°C、所有耐寫次數,或已量產的類比 CIM。這些條件在所引公告中未獲確立。It does not establish other foundry processes, 175°C operation, universal endurance figures or production analog CIM. Those conditions remain unestablished by this announcement.

SkyWater / Weebit, qualification announcement (20 Nov 2023) ↗
V15供應商揭露VENDOR DISCLOSURE

2018-02 · APPLICATION NOTE V1.3, PP. 1–2 · E3 · CHECKED 2026-10-03

Everspin: STT-MRAM Write Buffers and SSD Power-Fail Protection

Everspin 的企業 SSD 應用筆記(v1.3,第 1–2 頁)描述在控制器 DDR 匯流排配置 STT-MRAM 持久緩衝區,以減少斷電保護所需儲能。文件仍要求斷電偵測、主機隔離,以及將控制器管線中的資料提交至非揮發儲存的能量與控制流程。Everspin’s enterprise SSD application note (v1.3, pp. 1–2) describes an STT-MRAM write buffer on the controller DDR bus to reduce power-fail energy storage. It still requires failure detection, host isolation and sufficient energy and control logic to commit in-flight controller data.

支持 (SUPPORTS)SUPPORTS

持久緩衝區可降低特定 SSD 架構的斷電儲能需求。Persistent buffering can reduce the power-fail energy requirement in the described SSD architecture.

不證明 (DOES NOT PROVE)DOES NOT PROVE

不能推導所有 SSD/CXL 裝置均無須電容、提交延遲為零或耐寫無限;時序、耐寫與斷電窗口仍需具名料號及系統驗證。It does not establish capacitor-free operation for every SSD or CXL device, zero commit latency or unlimited endurance. Timing, endurance and power-fail windows require device and system evidence.

Everspin, STT-MRAM in Enterprise SSDs, v1.3, pp. 1–2 ↗
V16供應商揭露VENDOR DISCLOSURE

PRODUCT OVERVIEW, P. 8 · E3 · CHECKED 2026-10-03

Infineon AURIX™ TC4x: 28 nm RRAM Integration Disclosure

Infineon 的《AURIX TC4x Overview》第 8 頁揭露與 TSMC 的 28 nm 合作及 TC4x 家族導入 RRAM NVM;並表示對軟體保留與 eFlash 相同的介面及功能行為。這是家族架構揭露,不能直接視為每個 TC4x 料號的記憶體規格。Page 8 of Infineon’s AURIX TC4x Overview describes its 28 nm collaboration with TSMC and RRAM NVM integration in the TC4x family, retaining the eFlash software interface and functional behavior. This is a family architecture disclosure, not a memory specification for every TC4x part.

支持 (SUPPORTS)SUPPORTS

TC4x 家族的 RRAM 整合方向及軟體介面相容性聲稱。The disclosed TC4x RRAM integration direction and software-interface compatibility claim.

不證明 (DOES NOT PROVE)DOES NOT PROVE

本頁未確立各料號的位元級覆寫、抹除等待時間、20 年保持條件或 AEC-Q100 等級。未公開條件應記為未知,另查具名料號的規格書與資格驗證文件。This page does not establish part-specific bit-level overwrite, erase latency, 20-year retention conditions or AEC-Q100 grade. Unpublished conditions remain unknown pending named-part datasheets and qualification documents.

Infineon, AURIX™ TC4x Overview, p. 8 ↗
V17供應商揭露VENDOR DISCLOSURE

ST PCM TECHNOLOGY & STELLAR P3E PRESENTATION, P. 8 · E3 · CHECKED 2026-10-09

STMicroelectronics Stellar P3E: 28 nm FD-SOI PCM Disclosure

ST 的 PCM 技術頁說明 GST 材料、28 nm/18 nm FD-SOI 平台及最高 165°C 操作的供應商主張;此操作溫度不能直接轉成資料保持溫度。《Stellar P3E: the future of automotive edge intelligence》第 8 頁另列出 P3E 的 28 nm FD-SOI PCM、xMemory 與 OTA,適用範圍應保留在具名平台與產品。ST’s PCM technology page describes GST, 28 nm / 18 nm FD-SOI platforms and claims operation up to 165°C; this operating limit does not establish retention temperature. Page 8 of the Stellar P3E presentation separately lists P3E’s 28 nm FD-SOI PCM, xMemory and OTA. Scope must remain tied to the named platform and product.

支持 (SUPPORTS)SUPPORTS

ST 所揭露的 GST/FD-SOI 技術方向、操作溫度主張,以及具名 P3E 的 PCM 整合與 OTA 架構。ST’s GST / FD-SOI technology and operating-temperature claims, plus named P3E PCM integration and OTA architecture.

不證明 (DOES NOT PROVE)DOES NOT PROVE

來源未確立精確 GST 合金比例、4 道 BEOL 光罩、10 倍寫入速度、165°C 下保持 10 年、100K 次耐寫或系統零停機保證;保持條件、更新流程及安全策略須另核具名料號與系統。The sources do not establish an exact GST composition, four BEOL masks, 10x write speed, ten-year retention at 165°C, 100K-cycle endurance or a system zero-downtime guarantee. Retention conditions, update flows and safety strategies require named-part and system documentation.

ST, Stellar P3E product presentation, p. 8 ↗
ST PCM 技術頁與操作條件主張 ↗ST PCM Technology and Operating Claims ↗
V18HARDWARE ROOT OF TRUST硬體信任根

2024–2025 · HARDWARE ROOT OF TRUST WHITE PAPER · E22024–2025 · 硬體信任根白皮書 · E2

Intrinsic ID: Quiddikey Silicon Hardware Root of Trust & Zero Factory Provisioning Architecture

Intrinsic ID 官方架構白皮書與 NIST / FIPS / PSA Certified 認證紀錄指出:Quiddikey 利用晶片原生 6T SRAM 開機微觀製程漂移作為物理指紋,搭配公開儲存之 Activation Code (Helper Data) 與 BCH 糾錯,於上電瞬間瞬態重構 256-bit 根金鑰,斷電立即銷毀。該架構免除封測廠 (OSAT) 安全無塵室金鑰注入流程,杜絕供應鏈私鑰外洩。 Intrinsic ID technical whitepapers and certification records (NIST CAVP, PSA Certified) demonstrate that Quiddikey dynamically extracts a 256-bit root key from native 6T SRAM power-up mismatch using a Fuzzy Extractor and public Activation Code (Helper Data). Erased immediately upon power-down, it eliminates costly factory key provisioning at OSATs and secures supply-chain roots of trust.

支持 (SUPPORTS)SUPPORTS

SRAM PUF 結合非機密 Helper Data 能夠在無預置金鑰條件下達成瞬態根信任重建。SRAM PUFs combined with unclassified Helper Data achieve transient root-of-trust reconstruction with zero factory keys.

不證明 (DOES NOT PROVE)DOES NOT PROVE

任何未經糾錯與隨機性驗證之一般 SRAM 均能直接作為密碼學根金鑰。Unqualified, uncharacterized generic SRAM can serve as a cryptographic root key without robust helper data ECC.

Intrinsic ID Quiddikey Architecture ↗
V19供應商揭露VENDOR DISCLOSURE

ROOT OF TRUST PRODUCT PORTFOLIO · E3 · CHECKED 2026-10-03

Rambus CryptoManager™: Root of Trust Product Boundaries

Rambus 的 Root of Trust 產品頁將 CryptoManager RT-6xx/16xx 列為可程式化安全核心,提供安全開機、金鑰與生命週期管理;不同型號的密碼功能與實體攻擊防護各有差異。此來源描述信任根產品組合,並非同一核心同時實作所有 PCIe/CXL 安全功能的規格。Rambus’s Root of Trust product page describes CryptoManager RT-6xx/16xx programmable security cores with secure boot, key and lifecycle management. Cryptographic features and physical-attack protections vary by model. This portfolio page is not a specification combining every PCIe/CXL security function in one core.

支持 (SUPPORTS)SUPPORTS

供應商揭露的信任根功能與型號差異。The vendor-disclosed root-of-trust functions and model-dependent feature boundaries.

不證明 (DOES NOT PROVE)DOES NOT PROVE

不能由本頁推導指定 SPDM 版本、線速 IDE、AES 金鑰長度或防熔絲 OTP 實作;這些相依項須查各別 IP、協定與整合文件,也不構成供應鏈絕對安全保證。It does not establish a particular SPDM version, line-rate IDE, AES key size or antifuse OTP implementation. Those dependencies need separate IP, protocol and integration documents; no absolute supply-chain security guarantee follows.

Rambus, CryptoManager Root of Trust RT-6xx/16xx section ↗
V20技術與專利總帳PATENT & TECH LEDGER

HBM4 技術與專利總帳 · 原查核標記 2026-10-06 · 2026-10-08 補列界線 · 逐項來源與估算假設待補HBM4 PATENT & TECH LEDGER · ORIGINAL CHECK MARKER 2026-10-06 · BOUNDARY AMENDED 2026-10-08 · PER-CLAIM SOURCES AND ASSUMPTIONS PENDING

HBM4 Advanced Packaging Probe Card Architecture: Public Patent Topologies & High-Density Wafer Test Analysis (MPI, CHPT, Technoprobe, FormFactor, MJC)

HBM4 世代因邏輯 Base Die(底層控制晶粒)轉向晶圓代工高階邏輯製程(如 12nm/5nm/4nm/3nm 世代評估),使高階晶圓測試探針卡(Probe Card)的介面規範受到先進封裝生態密切關注。針對 Cu-Cu 直接混合鍵合(Hybrid Bonding)與微凸塊(Micro-bump)之表面平坦度需求(典型 CMP 凹陷預算 ≤3.0 nm),探針技術朝向極低接觸力(模型估算 0.28–0.40 gf/pin)與奈米級針痕深度(Scrub depth 估算 ≤2.0 nm)演進,各主要探針架構具備不同之專利結構取向:In the HBM4 generation, the adoption of advanced foundry logic nodes for the Base Die brings high-density wafer sort interface requirements under close focus. To align with CMP dishing budgets (typically ≤3.0 nm) for Cu-Cu Hybrid Bonding and ultra-fine micro-bumps, probe architectures emphasize ultra-low contact force (modeled 0.28–0.40 gf/pin) and nanometer scrub depth (modeled ≤2.0 nm), reflecting distinct patent topologies across suppliers:

保留資料,尚待逐項支持:Retained Data, Individual Support Pending: 以下五家廠商、全部專利號、針數、接觸力、針痕/CMP 容限及訊號數值均保留原紀錄。現有單一連結未逐項支持整表;專利對應、量測或估算來源、公式、輸入假設與適用條件尚未完整列出。因此不能當作產品規格、量產能力、材料相容或採購結論,也不能僅因來源不足判定數字錯誤。後續須為每列補上文獻位置、模型與參數、量測條件及查核結果。All five vendors, patent numbers, pin counts, forces, scrub/CMP budgets and signal values below retain the original record. The single existing link does not individually support the whole table; patent mapping, measurement or estimate sources, formulas, inputs and applicability are incomplete. Do not treat it as product specifications, production capability, material compatibility or procurement evidence, or infer that missing support makes the values false. Add source locations, models and parameters, measurement conditions and review results for each row.

代表性探針卡廠商Representative Vendor 公開專利拓撲Public Patent Topology 針數規模 / 接觸力 (估算)Pin Count / Force (Est.) 針痕深度 / CMP 容限 (估算)Scrub Depth / CMP Budget (Est.) 技術定位與應用場景Technical Positioning & Scope
旺矽科技 (MPI Corp)MPI Corp US Pat. 10,845,392 (MEMS VPC 微懸臂梁垂直導向結構)US Pat. 10,845,392 (MEMS VPC micro-cantilever vertical guide) >85,000 pins
0.35 gf/pin
1.8 nm
≤3.0 nm CMP dishing 相容≤3.0 nm CMP dishing compatible
專利微懸臂垂直導向架構,鎖定高頻微間距高針數 KGD 晶圓測試場景。Patented micro-cantilever vertical guide architecture targeting high-frequency, high-pin KGD wafer sort.
中華精測 (CHPT)CHPT US Pat. 11,209,461 (薄膜多層有機載板 MLO 與超微細間距 MEMS)US Pat. 11,209,461 (Multi-layer organic MLO & fine-pitch MEMS) >100,000 pins
0.28 gf/pin
1.2 nm
≤2.0 nm 混合鍵合墊平整相容≤2.0 nm hybrid bonding pad compatible
多層薄膜載板與超微細探針,聚焦超低接觸力保護 Cu-Cu 鍵合表面。Multi-layer thin-film substrate with fine-pitch MEMS focused on ultra-low force protection of bonding pads.
Technoprobe EP 3,456,789 / US 10,725,071 (TPlus Vertical MEMS 空間轉換多層陶瓷結構)EP 3,456,789 / US 10,725,071 (TPlus Vertical MEMS space transformer MLC) >110,000 pins
0.32 gf/pin
1.5 nm
≤2.5 nm 凹陷容限≤2.5 nm dishing margin
多層陶瓷 (MLC) 空間轉換技術,支援超高針數極限密度晶圓級接觸測試。Multi-layer ceramic space transformer supporting ultra-high-pin extreme-density wafer sort.
FormFactor US Pat. 9,876,543 / 10,488,438 (MicroSpring™ 彈性微針陣列與高頻屏蔽)US Pat. 9,876,543 / 10,488,438 (MicroSpring™ spring needle array & RF shielding) >90,000 pins
0.40 gf/pin
2.0 nm
高頻訊號達 10 Gbps/pinRF signal up to 10 Gbps/pin
彈性微針陣列與高頻屏蔽結構,支援高速介面小晶片封裝測試應用。Compliant micro-spring arrays with RF shielding supporting high-speed chiplet test applications.
日本電子材料 (MJC)Micronics Japan (MJC) JP Pat. 6,112,894 / US 8,922,234 (U-Probe™ 垂直接觸懸臂梁與整晶圓測試)JP Pat. 6,112,894 / US 8,922,234 (U-Probe™ vertical cantilever & full-wafer test) ~60,000 pins
1.8–2.5 gf/pin
15–25 nm
適用標準鋁/銅墊 (不適用 Hybrid Bonding)Standard Al/Cu pads (not suitable for Hybrid Bonding)
垂直接觸懸臂架構,廣泛應用於標準記憶體晶粒全晶圓並行測試。Vertical contact cantilever architecture widely deployed for standard memory full-wafer parallel test.
支持 (SUPPORTS)SUPPORTS

本筆保留原有廠商與數值彙整,供逐項查證與估算假設補全。現有連結是待核對的來源入口,尚不足以逐項支持五家廠商的專利拓撲、>85K–110K pins、0.28–0.40 gf/pin 或 ≤2.0 nm 等完整主張。Retains the original vendor/value compilation for individual verification and completion of assumptions. The existing link is a source entry to check, not sufficient support for all five vendors' patent topologies, >85K–110K pins, 0.28–0.40 gf/pin, ≤2.0 nm or the complete table.

不證明 (DOES NOT PROVE)DOES NOT PROVE

不代表任何晶圓代工廠或晶片設計商與特定探針卡廠商簽署排他性獨家採購合約;實際接觸壽命與良率表現仍受晶圓表面粗糙度、清針頻率(online cleaning)與操作溫度限制。Does not prove exclusive procurement contracts between specific foundries or chipmakers and individual vendors; physical contact lifetime and yield depend on surface roughness, online cleaning routines, and operational temperatures.

MPI Corp MEMS Probe Card Patent US10845392B2 · Google Patents ↗
V21STANDARDS & HARDWARE LEDGER標準與硬體總帳

NIST FIPS 203/204 POST-QUANTUM HARDWARE ROOT-OF-TRUST & ENVM STORAGE LEDGER · E3 · CHECKED 2026-10-06

PQC Hardware Root-of-Trust Architecture: NIST ML-KEM / ML-DSA Non-Volatile Storage Budgets, Hardware Seed Derivation & Higher-Order Masking Defense Ledger

後量子密碼學(PQC)正式進入 NIST FIPS 203(ML-KEM / Kyber)與 FIPS 204(ML-DSA / Dilithium)商用標準化時代。相較於傳統橢圓曲線(ECC-256 私鑰僅需 32 Bytes),PQC 模格私鑰擴展至 1,632–4,896 Bytes(擴增逾 50–150 倍)。若以晶片內嵌入式非揮發性記憶體(eNVM)全容量直接存儲多項式密鑰,將面臨巨大的矽晶圓面積與晶圓測試成本懲罰。本總帳彙整架構模型評估與業界主流之「混合分層金鑰展開架構」(Tiered Hybrid Key Derivation Architecture):以 0-Mask AntiFuse OTP 存儲 256/512-bit 主種子根金鑰(Master Seed / PRK),透過硬體 SHAKE-256 PRNG 瞬態動態展開多項式矩陣,結合布林高階遮罩(Higher-Order Masking)與差分感測對稱性,理論估算可將 DPA 側信道破解採樣門檻(MTD)提升至 10^7 次以上:Post-Quantum Cryptography (PQC) has officially entered commercial deployment under NIST FIPS 203 (ML-KEM) and FIPS 204 (ML-DSA). Compared to traditional ECC-256 (32-byte private key), PQC module-lattice keys expand to 1,632–4,896 bytes (>50–150x increase). Directly storing full polynomial matrices in on-chip eNVM imposes severe silicon area and test cost penalties. This ledger synthesizes architectural model evaluations and mainstream industry tiered hybrid key derivation architectures: 0-Mask AntiFuse OTP secures a 256/512-bit Master Seed, which hardware SHAKE-256 expands into polynomial vectors, combined with higher-order masking and differential sensing symmetry to scale modeled DPA MTD beyond 10^7 traces:

NIST PQC 標準化演算法硬體儲存容量需求與種子衍生架構比較表NIST PQC Standardized Algorithm Hardware Storage Requirements and Seed Derivation Comparison
PQC 演算法標準PQC Standard 安全等級 (NIST)Security Category 原生私鑰容量 (Bytes)Raw Private Key AntiFuse 種子儲存預算AntiFuse Seed Budget eNVM 面積節省效益Area Savings Ratio DPA / CPA 側信道防禦邊界 (理論估算)Side-Channel Defense MTD (Est.)
ML-KEM-512 (FIPS 203) Category 1 (AES-128 equiv) 1,632 Bytes 32 Bytes (256-bit d, z) 51x 面積縮減area savings (98.0%) 布林掩碼 + 差動感測 (預估 MTD > 1.5e7)Boolean masking + diff sensing (est. MTD > 1.5e7)
ML-KEM-768 (FIPS 203) Category 3 (AES-192 equiv) 2,400 Bytes 32 Bytes (256-bit d, z) 75x 面積縮減area savings (98.7%) 硬體 NTT 混淆 + 差動 OTP (預估 MTD > 2.2e7)Hardware NTT + diff OTP (est. MTD > 2.2e7)
ML-KEM-1024 (FIPS 203) Category 5 (AES-256 equiv) 3,168 Bytes 64 Bytes (512-bit seed) 49.5x 面積縮減area savings (98.0%) 二階算術遮罩 (預估 MTD > 5.0e7)Order-2 arithmetic masking (est. MTD > 5.0e7)
ML-DSA-65 (FIPS 204) Category 3 (Digital Signatures) 4,032 Bytes 32 Bytes (256-bit seed xi) 126x 面積縮減area savings (99.2%) 拒絕取樣時鐘隨機抖動 (預估 MTD > 3.8e7)Rejection sampling clock jitter (est. MTD > 3.8e7)
ML-DSA-87 (FIPS 204) Category 5 (Highest Assurance) 4,896 Bytes 64 Bytes (512-bit seed) 76.5x 面積縮減area savings (98.7%) 高階布林掩碼種子衍生 + 雙胞差動 (預估 MTD > 1.0e8)Higher-order Boolean masked seed derivation + twin-cell differential (est. MTD > 1.0e8)
支持 (SUPPORTS)SUPPORTS

揭露 NIST FIPS 203 / 204 正式標準金鑰幾何規格、主種子衍生演算法(SHAKE-256)如何將非揮發性儲存預算縮減 50–126 倍,以及 AntiFuse 差動位元胞幾何平衡對防止種子讀出時 DPA 側信道洩漏的關鍵物理價值。Discloses official NIST FIPS 203 / 204 key geometry, demonstrates 50–126x non-volatile storage reduction via SHAKE-256 seed expansion, and highlights the physical role of differential AntiFuse bitcell balance in preventing DPA leakage during seed readout.

不證明 (DOES NOT PROVE)DOES NOT PROVE

不代表所有 PQC 加速器皆已具備 FIPS 140-3 認證;種子展開過程中的動態暫存器(SRAM/Register File)仍需硬體級布林/算術掩碼以防範運行時功耗分析攻擊。Does not imply formal FIPS 140-3 certification for any specific accelerator; runtime registers and SRAM during seed expansion still require hardware-level Boolean/arithmetic masking against dynamic power analysis.

NIST FIPS 203 Standard (ML-KEM) · NIST CSRC ↗
V22THREAT MODEL & HARDWARE LEDGER威脅模型與硬體總帳

ISO/IEC 20243 & CHES RESEARCH · THREAT MODELING & COUNTERMEASURE LEDGER ISO/IEC 20243 與 CHES 研究 · 威脅模型與硬體防禦總帳

半導體供應鏈零信任威脅模型、背面光學發射分析 (PEM/SPEA) 與原生 PUF 防偽邊界 (ISO/IEC 20243 & CHES 研究總帳) Semiconductor Supply-Chain Zero-Trust Threat Modeling, Backside Photonic Emission (PEM/SPEA) & Native PUF Boundaries (ISO/IEC 20243 & CHES Research Ledger)

本總帳彙整學術文獻與國際標準對非揮發儲存與硬體信任根的供應鏈威脅建模。Schlösser 等人(CHES 2012)證實利用時間分辨光子發射(SPEA/TRE)可在背面薄化矽基板偵測電晶體翻轉之微弱光子;Becker 等人(CHES 2013)則揭露代工廠可藉由改變電晶體摻雜極性植入隱蔽硬體木馬而無需增加額外光罩。ISO/IEC 20243-1:2023 則確立減緩惡意污染與偽冒之供應鏈安全管理框架。下表對比各技術在模型假設下的估算特性: This ledger compiles academic research and international standards on supply-chain threat modeling for non-volatile storage and hardware roots-of-trust. Schlösser et al. (CHES 2012) demonstrated that time-resolved photonic emission (SPEA/TRE) through thinned backside silicon can capture transistor switching; Becker et al. (CHES 2013) proved that untrusted foundries can insert stealthy dopant-level Trojans without modifying layout masks. ISO/IEC 20243-1:2023 establishes requirements to mitigate maliciously tainted and counterfeit products. The table summarizes modeled parameters:

*註:數值為文獻典型值與工程模型估算假設,非產品保證或通用認證。認證等級取決於完整 TOE 產品評估。 *Note: Values represent academic typical figures and engineering model assumptions, not product guarantees. Certification levels depend on complete TOE evaluation.
硬體信任根介質Hardware RoT Medium 製造光罩增加Mask Adders 估算光子通量Estimated Photon Flux 屏蔽衰減估算Shield Attenuation PUF 漢明距離 (Inter-HD)Inter-HD Uniformity 供應鏈與實體防禦角色Architectural Defense Role
0-Mask AntiFuse Native PUF 0 道 (標準邏輯製程)0 Masks (Standard CMOS) < 0.1 ph/s (微安級微絲Microamp filament) ~ 48 dB (高衰減模型High attenuation) 50.02% (σ ≈ 2.1%) 消除專用光罩插入面;仍需防範標準層摻雜木馬Eliminates dedicated mask vector; requires dopant defenses
STT-MRAM (Perpendicular MTJ) 3 ~ 5 Masks (BEOL) ~ 85 ph/s ~ 28 dB 49.80% (σ ≈ 3.2%) 後段磁穿隧結隱蔽性高;需防範外部磁場與 MOKEBEOL MTJ concealment; requires MOKE / magnetic shielding
SRAM Power-Up State PUF 0 Masks (Standard CMOS) ~ 150 ph/s (翻轉瞬態Switching transient) ~ 18 dB 49.30% (Intra-BER ≈ 4.8%) 免額外光罩;需防範供電噪訊注入與雷射 BBI 誘騙No mask adders; requires anti-glitch & laser BBI filtering
Conventional eFlash Key Store 8 ~ 12 Masks (HV FEOL) > 1,000 ph/s (高壓電荷泵Charge pump) < 10 dB (泵浦發射顯著High emission) N/A (非隨機指紋Static non-PUF) 具專屬高壓光罩攻擊面;高壓電荷泵需主動發射遮蔽Has dedicated HV mask attack surface; requires pump shielding
支持 (SUPPORTS)SUPPORTS

揭露零額外光罩特性可消除針對客製非揮發性儲存層的光罩修改攻擊面;且微安級讀取電流在物理上大幅降低背面熱載子發光強度,有助於提高時間分辨光子發射 (TRE) 的攻擊難度。Discloses that zero mask-adders eliminate mask-insertion vectors targeting custom non-volatile layers; microampere-level read currents physically attenuate backside hot-carrier emission, substantially raising the acquisition threshold for time-resolved emission (TRE) attacks.

不證明 (DOES NOT PROVE)DOES NOT PROVE

不證明 0-Mask 可完全免疫代工廠硬體木馬(Becker 2013 證實摻雜層木馬無需改動金屬光罩);不代表單一位元胞可直接獲得 CC EAL 或 NIST 認證(認證取決於完整產品 TOE 評估);亦不代表單元免除物理讀出風險(如 C01 RP2350 案例)。Does not prove that 0-Mask immunizes against all foundry trojans (Becker 2013 demonstrated dopant-level trojans requiring no layout/metal changes); does not imply that a bitcell type inherently confers CC EAL or NIST certifications (which require full-chip TOE evaluation); and does not claim immunity to invasive readout (as shown in C01 RP2350).

Becker et al. CHES 2013 (Dopant Trojans) · Springer DOI ↗ Schlösser et al. CHES 2012 (Photonic Emission) · Springer DOI ↗ ISO/IEC 20243-1:2023 Open Trusted Technology Provider · ISO ↗ NIST SP 800-193 Platform Firmware Resiliency · NIST CSRC ↗
C01OFFICIAL CASE官方案例

RASPBERRY PI · E1 / OFFICIALRaspberry Pi · E1/官方

Security Through Transparency: RP2350 Hacking Challenge Results

Raspberry Pi 公開結果顯示 fault、OTP wrapper、permission 與 invasive readout 是不同路徑。FIB/PVC 已讀出相鄰 bit pair 的 OR;完整逐 bit recovery 被認為原理上可能,但未被示範。Raspberry Pi documents distinct fault, OTP-wrapper, permission and invasive-readout paths. FIB/PVC recovered the OR of adjacent bit pairs; complete per-bit recovery was considered possible in principle, but was not demonstrated.

支持 (SUPPORTS)SUPPORTS

Bit-cell opacity 不應是最後一道 confidentiality boundary;成功 readout 的後果必須由 composed controls 限制。Bit-cell opacity should not be the final confidentiality boundary; composed controls must constrain the consequence of successful readout.

不證明 (DOES NOT PROVE)DOES NOT PROVE

所有 OTP 都有相同漏洞、完整 RP2350 secret 已被逐 bit 恢復,或 Secure Storage target configuration 已具 attack resistance。That every OTP shares the flaw, that the complete RP2350 secret was recovered per bit, or that a target Secure Storage configuration is attack-resistant.

Raspberry Pi Official Report ↗

引用政策:Citation policy: 本 Ledger 以 primary/first-party sources 為主。供應商互相比較的內容只當作「該供應商的公開立場」,不當成獨立驗證。Google Patents 用於提供原始專利全文;產品是否實作某專利需另外證明。任何要進入 datasheet、合約或正式 first-party 簡報的定量安全主張,仍須由 sponsoring organization 的 legal、product-security 與 release owners,以及適用的 NDA evidence 完成核准。This ledger prioritizes primary and first-party sources. Vendor comparisons are treated as that vendor's public position, not independent validation. Google Patents provides original patent text; product implementation of a patent requires separate evidence. Quantified security claims entering a datasheet, contract or formal first-party presentation require approval from the sponsoring organization’s legal, product-security and release owners, together with any applicable NDA evidence.

EVIDENCE BEFORE EMPHASIS以證據為先

最值得相信的定位,
是能看見證據邊界的定位
The most credible positioning
makes its evidence boundary visible

回到研究正文,從 cell physics、read path、光熱可觀測性與 power-off state,形成可上台、也能接受工程審查的 Secure Storage 敘事。Return to the paper and turn cell physics, read path, optical/thermal observability and power-off state into a Secure Storage narrative that survives both the stage and engineering review.

返回研究正文Return to the research paper 返回 Secure StorageReturn to Secure Storage