An industrial control security protection system and method based on a four-layer physical isolation hardware architecture and an entropy source dynamic scrambling pipeline.

By using a four-layer physical isolation hardware architecture and an entropy source dynamic scrambling pipeline, the problems of observable adjudication trigger conditions, inferable static thresholds, lack of clock domain binding between cross-chip supply and judgment, and insufficient permission constraints in hardware-based chemical control systems are solved, achieving nanosecond-level response and low false alarm rate security protection.

CN123113523APending Publication Date: 2026-11-17CHANGSHA ARK ARTIFICIAL INTELLIGENCE TECHNOLOGY DEVELOPMENT CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202611289259.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing hardware-based chemical control system solutions suffer from several drawbacks: the adjudication trigger conditions can be intercepted by software paths; static adjudication thresholds can be observed and inferred; cross-chip entropy source supply and security judgment lack a unified clock domain binding; the upper-level master control has insufficient authority constraints on the lower-level security units; and nanosecond-level pass-through response and low false alarm rate in industrial electromagnetic environments are difficult to achieve simultaneously.

Method used

It adopts a four-layer physically isolated hardware architecture, including a security control layer, a business preprocessing layer, an entropy source layer, and a main control layer. Each layer is implemented by an independent chip. Through the entropy source dynamic scrambling pipeline and the entropy modulation discrimination window, it realizes dynamic scrambling of the adjudication threshold, source synchronous clock structure, and unidirectional read-only channel, forming a protection system with unobservable adjudication trigger conditions, response path without passing through software, and reliable operation in strong electromagnetic environment.

Benefits of technology

It achieves randomization of the adjudication threshold, hardware isolation of the response path, and suppression of electromagnetic interference, ensuring nanosecond-level response and low false alarm rate, and meeting the hardware-level protection requirements of critical information infrastructure.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123113523A_ABST
    Figure CN123113523A_ABST
Patent Text Reader

Abstract

This invention discloses an industrial control security protection system and method based on a four-layer physically isolated hardware architecture and an entropy source dynamic scrambling pipeline. The system uses a physically true random entropy code refreshed step-by-step for threshold dynamic scrambling and hardware decision timing pipelines, making triggering conditions unobservable and decision-making independent of software. The security control layer, service preprocessing layer, entropy source layer, and main control layer are implemented by physically independent chips; the security control layer is a CPLD or FPGA with non-volatile configuration memory, and the whitelist, threshold, and circuit breaker triple-judgment rules are hardware-fixed and cannot be rewritten during runtime; the service preprocessing layer outputs trigger levels via an independent hard-wired trigger channel, without passing through a data bus; the entropy source layer refreshes the entropy code and dynamically scrambles the decision threshold step-by-step at a period not exceeding 10ns; entropy code sampling, scrambling, judgment, and circuit breaker output are executed in the same hardware pipeline, with the entire process not exceeding 150ns, and the circuit breaker signal transitions within 80ns via an independent physical pin.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of industrial control system security technology, specifically relating to an industrial control security protection system and method based on a four-layer physical isolation hardware architecture and dynamically scrambling the hardware adjudication threshold with a physically true random entropy code that is refreshed step by step. It is applicable to high-level protection scenarios for critical information infrastructure such as power substation measurement and control devices, oil and gas refining emergency shutdown systems (ESD), rail transit signal safety systems, and high-end manufacturing PLC main controllers. Background Technology

[0002] (1) Current technological status The security protection of industrial control systems for critical information infrastructure such as power, petrochemicals, rail transportation, and high-end manufacturing has long relied primarily on software solutions. Security policies are calculated and decided by the main control CPU and operating system, then executed by the hardware. Typical forms include industrial firewalls, whitelist auditing, intrusion detection, and, in recent years, anomaly detection and situational awareness using artificial intelligence technologies. However, with the upgrading of attack methods, the software path itself has become part of the attack surface. In scenarios with high security requirements, technical solutions have emerged that implement some security functions through hardware circuits. Currently, representative hardware solutions fall into three main categories: first, introducing programmable logic devices such as CPLDs and FPGAs into the security control link, using hardware circuits to execute security adjudication rules; second, placing security decision-making, monitoring, and entropy source functions on physically independent chips, forming a board-level multi-chip physically isolated architecture, with physical fuses as the final disposal method; and third, using true random number (containing quantum random number) generator chips for key generation, startup verification, or as random factors in multi-module voting. Overall, the above hardware solutions still have shortcomings in terms of protection against adjudication trigger conditions and operational reliability in strong electromagnetic environments.

[0003] (2) Problems with existing technology A. Existing hardware implementation solutions suffer from the following interrelated problems: First, trigger signals can be intercepted by software. In existing solutions, trigger signals that report anomalies from the service processing unit to the security control unit are mostly transmitted via the bus, and the software layer can intercept, block, or tamper with them.

[0004] Secondly, while the adjudication rules can be fixed, the adjudication threshold is static and can be observed and inferred. The OTP one-time fixed strategy combined with physical circuit breaking disclosed in earlier application CN202610634170.8, and the three-module physical isolation, two-out-of-three voting circuit breaking system disclosed in the applicant's earlier application 202610355875.6 (published, publication number CN122513111A), both have static adjudication thresholds. In the latter, the true random number only participates in the decision-making as a random factor in voting and does not affect the adjudication threshold itself. Attackers can infer the triggering conditions and implement circumvention through long-term observation and statistical analysis.

[0005] Third, the entropy source supply and security determination across chips lack timing binding under a unified clock domain. The entropy source chip and the adjudication chip each have independent clock sources—the three-layer hardware in the applicant's earlier application 202610355875.6 has its own independent clock and independent power supply—the existing solution does not provide a clock structure for the two to work together in the hardware pipeline, and there may still be software-accessible nodes in the complete path from entropy source sampling to fuse output.

[0006] Fourth, the physical access constraints of the upper-level master controller on the lower-level security units are insufficient. The PLC two-factor authentication scheme disclosed in the earlier application CN202610183786.8 is based on the authentication protocol level; the unidirectional isolation interface disclosed in the earlier application CN202610193741.9 is located between the internal logic domains of a single chip and is used for key protection; neither of them solves the problem of physical access isolation between independent chips at the board level, where "the lower-level security switch cannot be accessed after the upper layer is compromised".

[0007] Fifth, it is difficult to achieve both nanosecond-level pass-through response and low malfunction rate in industrial electromagnetic environments. The faster the hardware pass-through response, the more sensitive it is to transient interference such as EFT pulse bursts, and existing solutions lack circuit-level mechanisms to suppress such malfunctions.

[0008] B. Specific differences from the applicant's earlier application: Taking the aforementioned prior art 202610355875.6 (publication number CN122513111A) as the closest prior art, the specific differences between this application and it include at least the following seven aspects: ① The drafting methods are different. The claims of the earlier application specify the specific chip model (the decision-making layer FPGA uses Anlu PH1A90SBG484) and the process node (180nm) of the monitoring layer ASIC; the claims of this application do not specify any specific model or process node, but define each layer with a composite definition of "function + connection + permission", and the device selection space is covered by the dependent claims and the embodiments in the specification.

[0009] ② The physical properties of the trigger channels are different. The prior application did not limit the physical channel form and wiring constraints of the abnormal trigger signal; this application limits the independent hard-wired trigger channel - without passing through any data bus, directly connecting the pins of the two devices with dedicated level signal lines (signal wires and reference ground), and requires that the wiring spacing between the trigger channel and the data channel be no less than 2mm in the PCB layout, and that continuous grounded copper foil isolation strips be used to replace the spacing where the spacing is less than 2mm in some places.

[0010] ③ The entropy sources function differently. In the prior application, the TRNG true random number only participates in the two-out-of-three voting as a random factor, without limiting the refresh period; in this application, the entropy source layer refreshes the entropy code every tick at a period of no more than 10ns. The tick-by-tick entropy code directly applies dynamic scrambling to the decision threshold and is embedded in the decision process as the first stage of the pipeline.

[0011] ④ The timing constraints differ. The prior application did not specify the pipeline structure or its delay parameters; this application specifies that the entire five-stage pipeline of "entropy code sampling → scrambling → whitelist determination → threshold determination → circuit breaker output" must not exceed 150ns, and there are no software-accessible nodes in the signal path.

[0012] ⑤ The limitations of the physical layout at the board level are different. The prior application only describes the physical isolation between the three modules in general terms, without limiting the board-level layout structure; this application limits the isolation slots and continuous ground planes between chip areas of each layer, the physical spacing between chips, and all layers are arranged on the same PCB, with all interlayer interconnections being PCB copper foil traces and no external traces.

[0013] ⑥ The hierarchical structure differs from the physical chip. The prior application uses a three-layer architecture: FPGA decision layer, ASIC monitoring layer, and TRNG entropy source layer, with the monitoring layer being a custom ASIC. This application uses a four-layer architecture, adding a service preprocessing layer and a main control layer. Each layer is implemented by a physically independent physical chip, and all have commercially available products without the need for custom tape-out. It also explicitly excludes the implementation method of integrating security adjudication functions and service preprocessing functions into the same chip.

[0014] ⑦ Different upper-level permission constraints. The earlier application did not address the permission issues of the main control layer and upper layers over the lower layers; this application restricts the main control layer to receive status information only through a one-way read-only data channel, and there are no reverse signal lines on the PCB pointing from the main control layer to the lower layers.

[0015] The differences mentioned above can be compared in the table below: Comparison Dimensions Prior application CN122513111A This application Security adjudication mechanism Separation of powers, two-thirds majority vote triggers circuit breaker The whitelist, threshold, and circuit breaker rules are solidified at the security control layer and directly adjudicated without a voting process. Writing style The specification must clearly state the specific chip model (Anlu PH1A90SBG484) and the 180nm ASIC process. Model number and manufacturing process are not specified; instead, they are limited by a combination of function, connectivity, and access permissions. Chip level Three layers: FPGA decision layer + ASIC monitoring layer + TRNG entropy source layer Four layers: A business preprocessing layer and a main control layer are added; each layer is an independent physical chip, eliminating single-chip integration. Trigger Channel The physical channel form of the trigger signal is not specified. Independent hardwired trigger channel, without any data bus; wiring spacing ≥2mm, with grounded copper foil insulation tape used where insufficient spacing is required. Adjudication threshold protection Static threshold; the entropy source output is only used as a random factor for voting, and the refresh period is not limited. The decision threshold is dynamically scrambled using a per-shot entropy code of no more than 10ns, making the effective threshold unpredictable for each shot. Pipeline timing Undefined pipeline and delay index The entire five-stage pipeline takes ≤150ns and has no software-accessible nodes. Upper-level permissions The main control layer and its access constraints were not involved. The main control layer only has a one-way read-only reception state, and the PCB has no reverse signal line. Power supply isolation Overview of independent power supply for each floor Star power supply + independent LDOs for each layer + independent ferrite beads + cascaded power-on reset with power-good connectors Physical layout This describes the physical isolation between modules without specifying the board-level structure. Isolation trench ≥2mm with continuous ground plane coverage, chip spacing ≥5mm, single PCB copper foil traces, no external cabling. Industrial Electromagnetic Compatibility Not involved Internal R&D testing: EFT ±4kV, -40℃ to +85℃ temperature cycling, no false triggering observed (see Example 5). (3) Industry needs that urgently need to be addressed In scenarios such as power substation monitoring and control, ESD in oil and gas refining, rail transit signal safety, and PLC main control in high-end manufacturing, safety regulations require protective devices to possess non-bypassable hardware-level final disposal capabilities, verifiable response times, and operational reliability in strong electromagnetic environments. Therefore, providing an industrial control safety protection system that offers unobservable adjudication trigger conditions, a response path that does not involve any software, and balances nanosecond-level response with a low false alarm rate is a pressing technical need for these industries. Summary of the Invention

[0016] (a) Purpose of the invention To address the aforementioned technical problems, this invention provides an industrial control security protection system and method based on a four-layer physically isolated hardware architecture and an entropy source dynamic scrambling pipeline. The inventive concept is as follows: the entropy source layer's step-by-step entropy code is simultaneously applied to the adjudication threshold layer (dynamic scrambling) and the decision timing layer (source synchronous pipeline), making the adjudication triggering condition unobservable in the time dimension; the source synchronous clock structure eliminates cross-clock domains, ensuring that the entire decision-fuse path does not pass through any software-accessible nodes; and the entropy modulation discrimination window suppresses malfunctions caused by transient electromagnetic interference.

[0017] (II) Technical Solution The technical solution of this invention is as described in the claims. In general, the system adopts a four-layer physically isolated hardware architecture: the security control layer, service preprocessing layer, entropy source layer, and main control layer are each implemented by physically independent chips, each equipped with an independent power conversion unit. There is no shared address bus or data bus between any two layers. The power conversion unit is a voltage conversion circuit (such as a low-dropout linear regulator (LDO) or a DC-DC converter) independently configured for each functional layer, thus forming independent power supply domains for each functional layer. Under this overall architecture, the security control layer, as the only hardware unit in the system with the highest security adjudication authority, incorporates two key mechanisms: First, the entropy source dynamic scrambling pipeline—embedding the physically true random entropy code refreshed by the entropy source layer on a clock-by-clock basis into a single-clock-domain hardware pipeline of "entropy code sampling → threshold scrambling → whitelist judgment → threshold judgment → circuit breaker output," making the adjudication threshold unpredictable on a clock-by-clock basis, and ensuring that the entire judgment path, not exceeding 150ns, does not pass through any software; second, the entropy modulation discrimination window—using a random-length window modulated by the entropy code to distinguish the authenticity of trigger signals on the independent hardwired trigger channel, suppressing malfunctions caused by transient electromagnetic interference. The service preprocessing layer reports characteristic data via a trusted data channel and emergency anomalies via an independent hardwired trigger channel; the entropy source layer supplies entropy on a clock-by-clock basis; the main control layer receives status information only via a one-way read-only data channel. This forms an overall industrial control security protection scheme that ensures "unobservable adjudication trigger conditions, unaccessible response paths in software, and reliable operation under strong electromagnetic environments."

[0018] In this specification, "hardwired trigger channel" refers to a hardware triggering path that connects two device pins directly via dedicated level signal lines (signal wires and reference ground) without going through any data bus.

[0019] The specific structure and implementation of each level are as follows: Security Control Layer: A CPLD or FPGA with non-volatile configuration memory, operating immediately upon power-on without reloading configuration; internally embedded whitelist, threshold, and circuit breaker rules, written once during configuration and unchangeable during operation; features an independent physical circuit breaker signal output pin; and a dedicated clock output pin outputs a source synchronization reference clock 54 to the entropy source layer. The security control layer also incorporates two key mechanisms: Firstly, the entropy source dynamic scrambling pipeline: The security control layer executes in the local clock domain according to the five-stage pipeline sequence: "Entropy code sampling stage 101 (≤10ns) → Scrambling operation stage 102 (≤20ns) → Whitelist determination stage 103 (≤40ns) → Threshold determination stage 104 (≤30ns) → Fuse output drive stage 105 (≤50ns)," with a total delay of no more than 150ns. The whitelist determination stage 103 is performed by a parallel comparator group for single-cycle parallel matching. The entire path does not pass through any software-accessible registers, interrupt vectors, or bus forwarding nodes. The fuse signal completes the level transition within 80ns via an independent physical pin. Secondly, the entropy modulation discrimination window: The security control layer determines the discrimination window length (2 to 5 clock cycles) by specifying the bit segment of the current entropy code; the trigger level of the independent hardwired trigger channel must be continuously valid for the entire clock cycle of the discrimination window to be confirmed, and any signal with a level interruption in the window is judged as an interference pulse and discarded.

[0020] Business preprocessing layer: Report message characteristic data through trusted data channel 51; when an anomaly requiring emergency handling is detected, output a trigger signal through independent hard-wired trigger channel 52; the wiring distance between the two channels is not less than 2mm, or when the distance is less than 2mm, a continuous grounded copper foil isolation strip 61 is provided between the two channels.

[0021] Entropy Source Layer: Independent true random number generator chip; synchronously outputs entropy code (8-bit or 16-bit) and valid indicator bit at a period of no more than 10ns on the clock edge of the source synchronization reference clock; connected to the security control layer and the service preprocessing layer via independent differential lines 53 and 56 respectively.

[0022] Main control layer: Status logs are received only through unidirectional read-only data channel 55; there are no reverse signal lines on the PCB; the unidirectional channel is physically unidirectional by optocouplers, magnetic couplers or unidirectional buffers.

[0023] (III) Beneficial Effects Compared with existing technologies, this invention improves upon existing technologies in three aspects simultaneously: the unobservability of safety decisions, the software inaccessibility of response paths, and the operational reliability in industrial electromagnetic environments, through a combined mechanism of "entropy source scrambling decision threshold + source synchronization pipeline timing binding + entropy modulation discrimination window". Specific beneficial effects are as follows: 1. Unobservable triggering conditions: The adjudication threshold changes randomly with each entropy refresh edge (≤10ns). The triggering condition measured by the attacker within any observation window becomes invalid for the next time step. Long-term statistical observation cannot infer the effective threshold. 2. Decision path is inaccessible to software: The entire process from entropy code sampling to circuit breaker output is completed within a single hardware pipeline in the same clock domain, without passing through any software-accessible nodes. The failure of the upper-layer system does not affect the decision and circuit breaker. 3. Mechanism-level suppression of electromagnetic interference malfunctions: Dynamic scrambling causes the effective threshold to wander randomly on each beat. Electromagnetic interference must hit the random threshold on the current beat to cause a false alarm, and the probability is much lower than hitting the fixed threshold. The entropy modulation discrimination window further requires the trigger signal to maintain an effective level throughout the window of unpredictable length (2~5 beats). Transient interferences such as EFT pulse groups cannot meet this condition. Together, these two constitute the circuit mechanism that achieves both nanosecond-level direct response and low false alarm rate (see Example 5 for internal R&D testing). 4. Trigger signal cannot be shielded: The independent hardwired trigger channel does not pass through any data bus, and software cannot intercept or shield it; 5. Startup window closure: Power-good cascade reset ensures that the security control layer is ready before the business layer, eliminating the startup exposure window of "business running first, decision not yet awake".

[0024] It is particularly important to point out that there has long been a technical bias in the field of industrial control security that "nanosecond-level hardware response cannot simultaneously meet the requirements of industrial-grade false alarm rate." This invention binds the entropy code refresh of no more than 10ns per cycle with a pipeline delay of no more than 150ns. While achieving nanosecond-level pass-through response, precisely because the adjudication threshold is dynamically unpredictable, electromagnetic interference must precisely hit the effective threshold of the random walk of each cycle to cause a false alarm. The probability of this is much lower than the probability of interference accidentally hitting the trigger condition under a fixed threshold scheme, thus overcoming the aforementioned technical bias (see Example 5 for internal testing during R&D).

[0025] In summary, this invention addresses the five prior art problems described in the background section: it solves the problem of trigger signals being intercepted by software paths using an independent hardwired trigger channel; it solves the problem of static adjudication thresholds being observable and inferred by using dynamic entropy source scrambling on a per-frame basis; it solves the problem of the lack of a unified clock domain for cross-chip entropy source supply and security judgment by using a source-synchronous clock structure and single-pipeline timing binding; it solves the problem of upper-layer compromise allowing access to lower-layer security switches by using a unidirectional read-only channel and physical permission constraints without reverse signal lines; and it solves the problem of the difficulty in achieving both nanosecond-level pass-through response and low false alarm rate by using a circuit-level mechanism of dynamic scrambling and entropy modulation discrimination window. This meets the industry requirements of critical information infrastructures such as power, petrochemical, rail transportation, and high-end manufacturing for hardware-level, non-bypassable, verifiable low false alarm protection capabilities. Attached Figure Description

[0026] Figure 1 This is a diagram of the overall system architecture of the present invention (also serving as an abstract drawing). Figure 2 This is a diagram of the internal pipeline structure of the safety control layer; Figure 3 This is a schematic diagram of the dynamic scrambling principle for the adjudication threshold; Figure 4 This is a timing diagram of the entropy modulation discrimination window; Figure 5 This is a diagram of the dual physical channel PCB layout and isolation structure; Figure 6 This is a diagram of a four-layer power supply link and a well-cascaded power supply structure; Figure 7 This is a flowchart of industrial control system security protection methods; Figure 8 This is the complete timing diagram of the circuit breaker response.

[0027] Figure labeling: 10—Security control layer; 11—Fuse output trace; 20—Service preprocessing layer; 30—Entropy source layer; 40—Main control layer; 51—Trusted data channel; 52—Independent hardwired trigger channel; 53—Differential trace (from entropy source layer to security control layer); 54—Source synchronization reference clock; 55—One-way read-only channel; 56—Differential trace (from entropy source layer to service preprocessing layer); 60—Isolation slot; 61—Grounded copper foil isolation strip; 101—Entropy code sampling stage; 102—Scrambling operation stage; 103—Whitelist determination stage; 104—Threshold determination stage; 105—Fuse output drive stage; 106—Base threshold register; 107—Entropy code register; 108—Threshold table. Detailed Implementation

[0029] like Figure 1As shown, the industrial control security protection system of the present invention consists of four physically independent chips: a security control layer 10, a service preprocessing layer 20, an entropy source layer 30, and a main control layer 40. The service preprocessing layer 20 reports characteristic data to the security control layer 10 via a trusted data channel 51 and outputs a trigger level to the security control layer 10 via an independent hardwired trigger channel 52. The independent hardwired trigger channel 52 is a dedicated level signal line and does not pass through any data bus. The entropy source layer 30 outputs entropy codes and valid indicator bits to the security control layer 10 step-by-step via differential traces 53. The security control layer 10 drives the sampling clock of the entropy source layer 30 via a source synchronization reference clock 54, outputs status information to the main control layer 40 via a one-way read-only channel 55, and outputs a fuse-breaking signal via a fuse-breaking output trace 11 upon confirmation of triggering. The external interface area includes a power input, tamper detection, fuse-breaking output, and a debugging port. The tamper detection is connected to the tamper detection input terminal of the security control layer 10, and the debugging port is connected to the main control layer 40.

[0030] In this embodiment, the security control layer 10 uses a MachXO3D series security control chip (a non-volatile device with Flash architecture, which works upon power-up; the actual measured model of the prototype is MachXO3D-9400HC, and the scope of protection of the claims is not limited to a specific model), the service preprocessing layer 20 uses an FPGA chip (a general-purpose device, whose logic resources are sufficient for protocol parsing and feature extraction), the entropy source layer 30 uses an independent true random number generator chip (based on quantum tunneling effect, vacuum fluctuations or circuit noise (such as semiconductor thermal noise) entropy source; the entropy code width is 8 bits or 16 bits, and the refresh period is no more than 10ns; the required entropy output rate can be achieved by parallel synthesis of multiple entropy sources on the chip), and the main control layer 40 uses an RK3568 series main control processor chip (the actual measured model of the prototype is RK3568B2, and the RK3568J of the same series can also meet the functional requirements of this invention; the scope of protection of the claims is not limited to this model). The four chips are arranged on the same PCB, each equipped with an independent LDO (Power Direction Controller). The power supply circuit is led out separately from the common power node in a star topology and connected in series with a ferrite bead (e.g., Figure 6 (As shown); a 2mm wide isolation trench 60 is provided between each chip area and covered with a continuous ground plane, and the physical spacing between chips is not less than 5mm (e.g. Figure 5 (As shown); the physical fuse signal output pin of the safety control layer 10 is led out to the external interface area through the fuse output trace 11. The width of the trace is not less than 1mm, and both sides are covered with ground (accompanied by grounded copper foil).

[0031] Source-synchronous clock structure: The on-chip clock of the security control layer drives the sampling clock input of the entropy source layer chip via a dedicated clock output pin and a point-to-point clock trace on the PCB (with the same length as the entropy code differential trace to control clock skew). The entropy source layer synchronously outputs a 16-bit entropy code and a 1-bit valid indicator bit (the valid indicator bit indicates that the entropy code for this cycle has passed the on-chip self-test) on each clock edge. These two bits, along with the reference clock, form a source-synchronous interface. The security control layer directly samples the entropy code and valid indicator bit in its local clock domain, updating the scrambling entropy code register only when the valid indicator bit is valid. Therefore, the entropy source supply and circuit breaker determination are in the same clock domain, and there is no asynchronous cross-clock domain handshake between pipeline stages. The timing closure of the "entropy code sampling—determination—circuit breaker" path can be fully verified through static timing analysis.

[0032] It should be noted that the functional requirements of each layer of devices in this invention are common capabilities of the corresponding device categories: the non-volatile configuration storage, power-on operation, and configuration locking required by the security control layer are common attributes of CPLDs and Flash architecture FPGAs, and not specific to any particular model; the per-frame refresh and output rate indicators required by the entropy source layer are common indicators that can be achieved by true random number generator chips based on quantum tunneling effect, vacuum fluctuations, or circuit noise; the main control layer only requires general computing and log receiving capabilities. Therefore, those skilled in the art can implement the solution of this invention by selecting any commercially available device that meets the common capability requirements within the device categories defined in the claims without any inventive effort. In another embodiment, the security control layer can also be implemented using domestically produced programmable logic devices with the same function (such as CPLDs or Flash architecture FPGAs from AGM, Anlu, or Gowin), whose non-volatile configuration storage, power-on operation, and configuration locking capabilities, as well as the security adjudication logic resources required by this invention, can all be met. The security adjudication logic function is completely consistent, and there is no need to change the device selection and circuit structure of other layers.

[0033] Implementation Method 1 (XOR / Modal plus dithering): For example... Figure 2 and Figure 3 As shown, the security control layer includes a baseline threshold register 106 (the baseline value is written once during the configuration phase), an entropy code register 107, and a scrambling operation circuit (i.e., scrambling operation stage 102). At each entropy code refresh edge, the scrambling operation circuit performs a bitwise XOR operation (or modulo-dithering the lower-order bits of the entropy code) on the baseline threshold and the current 16-bit entropy code. The result is written to the current valid threshold register and used as the current valid decision threshold, which is then fed into the threshold comparison circuit (threshold determination stage 104). The comparison circuit compares the packet feature count value reported by the service preprocessing layer with the current valid decision threshold and outputs the determination result. The valid threshold observed by an attacker at any given time is a random walk value, and it becomes invalid in the next frame.

[0034] Implementation Method 2 (Entropy Code Index Threshold Table): A threshold table 108 containing 8 candidate thresholds (threshold levels 0-7) is pre-stored within the security control layer (written once during the configuration phase). At each entropy code refresh edge, the lower 3 bits of the current entropy code are used as an index to select the valid decision threshold for that current step. This method has lower circuit overhead and is suitable for devices with limited logic resources.

[0035] Both methods guarantee that the effective threshold update frequency is the same as the entropy refresh frequency (≤10ns), and the update process does not pass through any software writable registers.

[0036] Furthermore, before reporting feature data, the business preprocessing layer 20 can perform cross-domain whitelist pre-filtering on industrial control bus messages (pre-screening by protocol type, source / destination address, or instruction code to block obviously illegal messages). The whitelist judgment level 103 in the pipeline performs the final whitelist judgment, thus forming a heterogeneous dual-verification structure of "preprocessing layer pre-filtering + security control layer final judgment". It is important to distinguish that: the pre-filtering rules can be updated with business configuration, while the whitelist judgment rules, which serve as security adjudication rules, are fixed in the security control layer and cannot be rewritten by software during runtime; the two have different functions.

[0037] like Figure 4 As shown, the security control layer 10 performs step-by-step sampling of the level state of the independent hardwired trigger channel 52. The confirmation logic is as follows: the identification window length N (N∈{2,3,4,5} beats) is obtained by adding 2 to the lower 2 bits of the current beat entropy code; the trigger level must be valid for all N consecutive beats to be determined as a valid trigger and enter the fuse-breaking determination level; if the level fails in any beat within the window, the trigger is determined as an interference pulse and discarded, and the discard counter is incremented by one (for read-only query by the main control layer).

[0038] Since the window length is determined by the current entropy code, interference sources (including EFT pulse groups and RF field-induced noise) cannot predict the current window length; the probability of transient interference pulses maintaining an effective level throughout the unpredictable 2-5 beat window is extremely low. This mechanism, together with threshold randomization, enables the system to maintain a nanosecond-level direct response while achieving a low false alarm rate in industrial electromagnetic environments.

[0039] The complete timing sequence of the circuit breaker response is as follows: Figure 8 As shown: the whitelist determination and threshold determination are established first. The hard-wired trigger level is confirmed after being continuously valid throughout the identification window (N=3 in the figure). The fuse output level completes the transition within ΔT≤80ns after the trigger confirmation. The total delay from entropy code sampling to fuse signal output is no more than 150ns.

[0040] like Figure 1 and Figure 5As shown, the one-way read-only channel 55 (one-way read-only data channel) is used by the security control layer 10 to output status information (circuit break time, trigger source identifier, message feature digest, drop count, etc.) to the main control layer 40. Its physical one-wayness is guaranteed by any of the following methods: Optical isolation: Data signals flow from the security control layer side to the main control layer side via optical couplers. There is no reverse transmission path in the physical structure of the optical couplers, and electrical isolation is also achieved. Magnetic coupling isolation: A digital magnetic isolator is used, and the transmission direction is fixed by the device wiring; One-way buffer: It adopts a one-way buffer gate with a fixed transmission direction (the input end is connected to the security control layer and the output end is connected to the main control layer). There are no signal lines on the PCB that point from the main control layer to the security control layer or the service preprocessing layer.

[0041] All three implementations satisfy the condition that "the master control layer does not have a physical path to send any configuration instructions or control signals to the lower layer".

[0042] In addition, such as Figure 1 and Figure 5 As shown, the per-frame entropy code of the entropy source layer 30 is also simultaneously output to the service preprocessing layer 20 via an independent differential line 56, providing encryption seeds for its communication data and realizing multiple uses of a single source.

[0043] Power-good cascade power-on: e.g. Figure 6 As shown, after reverse connection protection and filtering, and main power conversion, the external power supply branches into four power supply branches in a star topology at the common power node (star node). Each branch is connected in series with a ferrite bead and regulated by an independent LDO at its own level. The power good (PG) indicator output of each LDO is connected to the safety control layer via the PG bus. Upon power-up, the safety control layer is powered on first. After completing the self-test of the solidified logic, it sequentially confirms the power good signals of the entropy source layer, the service preprocessing layer, and the main control layer, in the order of "entropy source layer → service preprocessing layer → main control layer". Figure 6 (Marked ①②③) release the reset step by step; when any layer's power good indicator fails, the security control layer retains or resets the reset of the remaining layers to ensure that the service channel is not opened before the decision layer is ready.

[0044] Anti-tamper linkage: Security control layer 10 is connected to the anti-tamper detection input terminal (chassis anti-tamper switch or PCB anti-tamper mesh trace). When the anti-tamper signal is valid, the solidified logic performs root key erasure within the same clock domain and drives the fuse output trace 11 to change, without going through the software path at all.

[0045] Industrial Electromagnetic Compatibility Verification: The R&D prototype underwent internal testing according to GB / T 17626.4 (Electrical Fast Transient Burst Immunity, Test Level ±8kV) and full-temperature cycling conditions (-40℃~+85℃). During the testing period, a total of 120,000 abnormal bus messages were injected. All level transition events of the fuse-blown output pin were recorded and compared one by one with the excitation log. No false triggering or missed interception occurred during the EFT burst application period and throughout the entire temperature cycling process. The above results are from internal R&D testing and are not from a third-party testing report.

[0046] like Figure 7 As shown in the figure, this embodiment illustrates the industrial control security protection method implemented using the above system, including the following steps: S1: When the system is powered on, the security control layer is powered on first and completes the initialization and self-test of the solidified logic. Then, according to the power good indication signal of each functional layer, the reset signals of the business preprocessing layer, entropy source layer and main control layer are released in a preset order. S2: The security control layer outputs a source synchronization reference clock to the entropy source layer through a dedicated clock output pin; the entropy source layer synchronously outputs the entropy code and valid indicator bit on each clock edge of the source synchronization reference clock at a period of no more than 10ns, and transmits them to the security control layer through independent differential traces. S3: The business preprocessing layer collects industrial control bus messages in real time, performs protocol parsing and abnormal behavior feature extraction, and reports key feature data to the security control layer through a trusted data channel; S4: The security control layer samples the entropy code in the local clock domain, uses the current entropy code to dynamically scramble the decision threshold, and performs entropy code sampling, threshold scrambling, whitelist determination, threshold determination, and fuse output determination in the same hardware pipeline according to the clock tick sequence; the total delay from entropy code sampling to fuse signal output is no more than 150ns, and the signal path of the hardware pipeline does not pass through any software-accessible registers, interrupt vectors, or bus forwarding nodes; S5: When the service preprocessing layer detects an anomaly requiring emergency handling, it changes the level state and outputs a trigger signal through an independent hard-wired trigger channel; the security control layer determines an identification window of 2 to 5 clock cycles in length using a specified bit segment of the current entropy code. The trigger signal is confirmed as a valid trigger when it remains valid for all clock cycles within the identification window. Subsequently, the security control layer completes the output level transition within 80ns through an independent physical fuse signal output pin. This transition is not confirmed by any software. S6: After the security control layer completes the output level transition, it outputs the status information, including the transition time, trigger source identifier, and message feature summary, to the main control layer through a one-way read-only data channel.

[0047] The correspondence between the above method flow and the method claim (claim 10) is as follows: S1 corresponds to good power supply cascading, S2 corresponds to source synchronization clock establishment, S4 corresponds to dynamic scrambling pipeline, and S5 corresponds to entropy modulation discrimination window confirmation and fuse output. Figure 7 The two decision nodes, "meeting the circuit breaker condition" and "remaining valid within the window", correspond to the circuit breaker trigger determination and entropy modulation identification window confirmation, respectively. If the determination result is negative, the process returns to continue monitoring or discards the trigger without entering the circuit breaker output.

[0048] In practice, there are architectures that integrate security adjudication functions and business preprocessing functions onto the same chip (such as a single multi-core processor chip or a single FPGA chip). This type of architecture differs substantially from the present invention: the adjudication logic and business logic are integrated within the same die, sharing an on-chip power network, failing to form an independent power supply domain and board-level physical isolation; the on-chip configurable paths also cannot exclude software-accessible nodes. Therefore, this type of architecture does not achieve the technical effects of the present invention and is not within the scope of protection of the claims of this invention.

[0049] The components in each layer of this invention can be selected according to functional requirements: the security control layer uses a programmable logic device with non-volatile configuration memory, power-on operation, and configuration locking support (the prototype was tested using the MachXO3D-9400HC security control chip); the main control layer uses a processor chip with general computing and log management capabilities (the prototype was tested using the RK3568B2 main control processor); the business preprocessing layer uses an FPGA chip with logic resources sufficient for industrial control protocol parsing and feature extraction; the entropy source layer uses a true random number generator chip with an entropy code width of 8 or 16 bits and a refresh period of no more than 10ns, and the equivalent entropy output rate can be met by parallel synthesis of multiple entropy sources on the chip. All of the above components are commercially available and do not require custom fabrication. The PCB isolation structure, pipeline, and discrimination window logic are all within the range achievable with mature board-level processes and small-to-medium scale logic resources, allowing for mass production and application in critical infrastructure scenarios such as power, petrochemical, rail transportation, and high-end manufacturing.

Claims

1. An industrial control security protection system based on a four-layer physically isolated hardware architecture, comprising four functional layers—a security control layer, a service preprocessing layer, an entropy source layer, and a main control layer—set on the same PCB. Each functional layer is implemented by a physically independent chip and is equipped with an independent power conversion unit. There is no shared address bus or data bus between any two functional layers; characterized in that: The security control layer is a complex programmable logic device (CPLD) or field-programmable gate array (FPGA) with non-volatile configuration memory, which operates upon power-up without loading configuration from external memory. The security decision logic inside the security control layer is fixed in hardware circuit form. The security decision logic includes message whitelist determination rules, threshold determination rules, and fuse trigger determination rules. These rules are written once during the hardware configuration phase and cannot be modified, overwritten, or bypassed by any software instructions during system operation. The security control layer has an independent physical fuse signal output pin. The service preprocessing layer and the security control layer are connected by two physically independent channels: a trusted data channel and an independent hardwired trigger channel. The independent hardwired trigger channel is a dedicated level signal line that does not pass through any data bus. When the service preprocessing layer detects an anomaly, it outputs a trigger signal by changing the level state of the dedicated level signal line. The entropy output terminal of the entropy source layer is connected to the security control layer through independent differential lines. The entropy source layer refreshes the output entropy code every time at a period of no more than 10ns. The security control layer uses the entropy code refreshed every time to dynamically scramble the adjudication threshold in the threshold determination rule, so that the adjudication threshold changes unpredictably in the time dimension. The security control layer outputs a source synchronization reference clock to the entropy source layer through a dedicated clock output pin. The entropy source layer synchronously outputs the entropy code and a valid indicator bit on the clock edge of the source synchronization reference clock. The security control layer samples the entropy code in the local clock domain. Entropy code sampling, threshold scrambling, whitelist determination, threshold determination, and fuse output are executed in the same hardware pipeline in the clock tick order of the local clock domain. The total delay from entropy code sampling to fuse signal output is no more than 150ns. The signal path of the hardware pipeline does not pass through any software-accessible registers, interrupt vectors, or bus forwarding nodes. When the fuse trigger determination rule is met, the security control layer completes the output level transition within 80ns through the physical fuse signal output pin, and the output level transition is not confirmed by any software.

2. The system according to claim 1, characterized in that, The message whitelist determination rules include: a whitelist of allowed industrial control protocol type identifiers, a whitelist of allowed source / destination addresses, and a whitelist of allowed industrial control instruction codes; the threshold determination rules include: a maximum threshold for the number of messages per unit time, an abnormal message length threshold, and an abnormal instruction execution frequency threshold; the circuit breaker trigger determination rules include: triggering upon whitelist matching failure, triggering upon threshold exceeding the limit, and triggering upon confirmation of the validity of the trigger signal from an independent hardwired trigger channel.

3. The system according to claim 1, characterized in that, The security control layer performs entropy modulation discrimination on the trigger signal output by the independent hardwired trigger channel: the security control layer determines the discrimination window length by a specified bit segment of the current clock entropy code, and the discrimination window length is 2 to 5 clock clocks; the level state of the trigger signal is confirmed as a valid trigger only when it is continuously valid throughout all clock clock clocks of the discrimination window; the trigger signal with a level interruption within the discrimination window is judged as an interference pulse and discarded.

4. The system according to claim 1, characterized in that, The source synchronization reference clock is transmitted to the sampling clock input of the entropy source layer via a point-to-point clock trace on the PCB and a dedicated clock output pin of the security control layer. The entropy source layer outputs an 8-bit or 16-bit entropy code and a 1-bit valid indicator bit every refresh cycle of no more than 10ns. The security control layer updates the entropy code used for dynamic scrambling only when the valid indicator bit is valid.

5. The system according to claim 1, characterized in that, The dynamic scrambling is implemented in one of the following two ways: Method 1: The security control layer is equipped with a reference threshold register and a scrambling operation circuit. The scrambling operation circuit performs a bitwise XOR operation or a modulo-jitter operation on the reference threshold in the reference threshold register and the current entropy code, and uses the operation result as the current valid decision threshold; Method 2: The security control layer pre-stores a threshold table containing multiple candidate decision thresholds, and selects the current valid decision threshold from the threshold table using a specified bit segment of the current entropy code as an index; The current valid decision threshold is updated once with each entropy code refresh edge.

6. The system according to claim 1, characterized in that, The business preprocessing layer is a field-programmable gate array (FPGA); the entropy source layer is a true random number generator chip based on quantum tunneling effect, vacuum fluctuations, or circuit noise; and the main control layer is a general-purpose processor chip based on ARM or RISC-V architecture.

7. The system according to claim 1, characterized in that, Only a one-way read-only data channel is provided between the main control layer and the security control layer. At the physical level, the one-way read-only data channel only allows data to flow from the security control layer to the main control layer. There are no reverse signal lines on the PCB pointing from the main control layer to the security control layer or the service preprocessing layer. The one-way read-only data channel is implemented by any one of an optocoupler isolation device, a magnetic coupler isolation device, or a one-way buffer with a fixed transmission direction. After the security control layer completes the output level transition, it outputs status information containing the transition time, trigger source identifier, and message feature digest through the one-way read-only data channel.

8. The system according to claim 1, characterized in that, Each of the aforementioned functional level power conversion units is equipped with a power good indication output terminal. When the system is powered on, the security control layer powers on first. After the security control layer completes the self-test of the solidified logic, it releases the reset signals of the remaining functional levels in a preset order according to the power good indication signals of each functional level. If the power good indication of any functional level fails, the security control layer maintains or resets the reset of the remaining functional levels. The security control layer is also connected to an anti-tamper detection input terminal. When the anti-tamper signal of the anti-tamper detection input terminal is valid, the solidified logic of the security control layer performs a root key erasure operation and simultaneously completes the output level transition through the physical fuse signal output pin. The root key erasure operation does not go through any software path.

9. The system according to claim 1, characterized in that, The hardware pipeline consists of an entropy code sampling stage, a scrambling operation stage, a whitelist determination stage, a threshold determination stage, and a circuit breaker output driver stage. The entropy code sampling stage has a latency of no more than 10 ns, the scrambling operation stage has a latency of no more than 20 ns, the whitelist determination stage has a latency of no more than 40 ns, the threshold determination stage has a latency of no more than 30 ns, and the circuit breaker output driver stage has a latency of no more than 50 ns. The whitelist determination stage uses a parallel comparator group to perform single-cycle parallel matching between the message feature data and the message whitelist determination rules.

10. An industrial control system security protection method, characterized in that, The system described in any one of claims 1 to 9 is used to perform the following steps: S1: When the system is powered on, the security control layer is powered on first and completes the initialization and self-test of the solidified logic. Then, according to the power good indication signal of each functional layer, the reset signals of the service preprocessing layer, the entropy source layer and the main control layer are released in a preset order. S2: The security control layer outputs a source synchronization reference clock to the entropy source layer through a dedicated clock output pin; the entropy source layer synchronously outputs the entropy code and valid indicator bit on each clock edge of the source synchronization reference clock at a period of no more than 10ns, and transmits them to the security control layer through independent differential traces. S3: The business preprocessing layer collects industrial control bus messages in real time, performs protocol parsing and abnormal behavior feature extraction, and reports key feature data to the security control layer through a trusted data channel. S4: The security control layer samples the entropy code in the local clock domain, dynamically scrambles the decision threshold using the current entropy code, and performs entropy code sampling, threshold scrambling, whitelist determination, threshold determination, and circuit breaker output determination in the same hardware pipeline according to the clock tick sequence; the total delay from entropy code sampling to circuit breaker signal output is no more than 150ns, and the signal path of the hardware pipeline does not pass through any software-accessible registers, interrupt vectors, or bus forwarding nodes; S5: When the service preprocessing layer detects an anomaly requiring emergency handling, it changes the level state and outputs a trigger signal through the independent hardwired trigger channel; the security control layer determines an identification window of 2 to 5 clock cycles in length using a specified bit segment of the current entropy code. The trigger signal is confirmed as a valid trigger when it remains valid for all clock cycles within the identification window. Subsequently, the security control layer completes the output level transition within 80ns through an independent physical fuse signal output pin. The output level transition is not confirmed by any software. S6: After the security control layer completes the output level transition, it outputs the status information, including the transition time, trigger source identifier and message feature digest, to the main control layer through a one-way read-only data channel.

Citation Information

Patent Citations

  • Security key distribution system, method and chip based on physical random entropy source

    CN122001573A

  • AI behavior full-link compliance detection and physical fusing method and device based on hardware root of trust

    CN122286729A

  • PLC instruction-level two-factor authentication system and method based on hardware root of trust

    CN122293358A

  • Hardware-level security isolation system based on FPGA + ASIC + TRNG

    CN122513111A