Trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF

Abstract

The present invention discloses a trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF, which includes a trusted image sensor, a bias adjustment circuit, a row scanner, a column scanner, a signal readout circuit, a control circuit, an analog selector and a comparator, wherein the trusted image sensor includes a plurality of photosensitive pixel units; the bias adjustment circuit is connected to each column of photosensitive pixel units; each row of photosensitive pixel units is connected to the row scanner; each column of photosensitive pixel units is connected to the signal readout circuit and the analog selector, the analog selector is connected to the comparator, and the comparator is used to output a PUF response; the control circuit is connected to the row scanner and the signal readout circuit to provide a first control signal; the column scanner is connected to the analog selector and the signal readout circuit to provide a second control signal for the analog selector and provide a column scanning signal for the readout circuit to realize digital imaging signal readout. The present invention realizes the trusted sensing function of the image sensor.

Description

Technical Field

[0001] The present invention relates to the technical field of integrated circuit hardware security, and in particular to a trusted image sensor chip based on a 4T-APS optically reconfigurable strong PUF. Background Art

[0002] In recent years, driven by industries such as smartphones, Internet of Things (IoT), Industry 4.0 and 5G, the Internet of Things represented by CMOS image sensors (CIS) has developed rapidly. According to the French Yole Group, the global market size of CMOS image sensors will reach US$29 billion by 2028. In addition, image sensor nodes are widely distributed in IoT edge devices and are used in scenarios involving a large amount of private information such as face payment, fingerprint recognition, identity authentication, confidential video conferencing and authorized digital views, which has led to an urgent need for information security for its applications.

[0003] As an important part of the IoT ecosystem that changes human lifestyles, image sensors have always been a weak link and target of cyber-physical attacks. According to a report from SAM, more than 1 billion attacks on the IoT occurred in 2021, of which nearly 900 million were IoT-related phishing attacks, and attacks on cameras alone reached 27 million. In addition, image sensors are also facing threats such as demonstration attacks, virtual camera attacks, and image forgery / tampering using generative artificial intelligence (AI) technology. In addition, due to the stringent requirements of image sensor devices on size, cost, and portability, the energy consumption and processing power of many image sensor devices are highly limited, making it difficult to meet high security requirements.

[0004] In view of the above security threats and the security vulnerabilities caused by the untrustworthiness of image sensor chips, how to implement anti-spoofing technology from the hardware level with image sensors as the core has attracted widespread attention from researchers in the field. In order to achieve the security authentication of the above image sensors, that is, to develop trusted image sensors, researchers have proposed a technical solution in recent years to integrate general encryption and authentication circuit modules into image sensors to ensure the security and authenticity of terminal sensor nodes, and upper-level users do not need to worry about the relatively complex software security issues in the imaging system. However, this solution has problems such as high cost, complex process, high power consumption, and the keys stored in non-volatile memory are vulnerable to physical intrusive attacks, which makes it not widely used in the field of image sensors. As a lightweight hardware security primitive, the Physically Unclonable Function (PUF) can easily generate data information with uniqueness, randomness, and unclonability by forward extracting the process deviations that are inevitably introduced during the chip manufacturing process. Moreover, this information is generated in real time and does not require additional storage, which saves hardware resources and reduces the risk of physical attacks. It can provide an economically feasible lightweight solution for key generation, identity authentication, and information security. Therefore, how to integrate image sensors with strong PUF security technology to realize the trusted sensing function of image sensors is an urgent problem to be solved.

[0005] Therefore, in order to solve the above problems, the present invention provides a trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF, which can integrate the image sensor with the strong PUF. Summary of the invention

[0006] The present invention provides a trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF, aiming to solve the problem that existing image sensors cannot realize trusted sensing functions.

[0007] In order to solve the above technical problems, the present invention provides a trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF, which includes: a trusted image sensor, a bias adjustment circuit, a row scanner, a column scanner, a signal readout circuit, a control circuit, an analog selector and a comparator, wherein the trusted image sensor includes a plurality of photosensitive pixel units, each of which can be independently photosensitive to output a corresponding photosensitive output voltage; the bias adjustment circuit is connected to each column of the photosensitive pixel units in the trusted image sensor to bias the input current in each column of the photosensitive pixel units. ; The control signal input end of each row of the photosensitive pixel units is connected to the control end of the row scanner; the output end of each column of the photosensitive pixel units is connected to the signal readout circuit and the analog selector, the analog selector is connected to the comparator, and the comparator is used to output the PUF response; the control circuit is connected to the row scanner and the signal readout circuit to provide a first control signal to the row scanner and the signal readout circuit; the column scanner is connected to the analog selector and the signal readout circuit to provide a second control signal to the analog selector and scan the digital imaging signal in the signal readout circuit by column.

[0008] Furthermore, the bias adjustment circuit includes a tail current source tube and a current mirror, and the tail current source tube and the current mirror are respectively connected to the voltage output end of each photosensitive pixel unit in a column of the photosensitive pixel units.

[0009] Furthermore, the drain of the tail current source tube is connected to the current mirror and the photosensitive pixel unit, the source thereof is grounded, and the gate thereof is used to input a reference voltage.

[0010] Furthermore, the photosensitive pixel unit includes a photodiode, a transfer transistor, a reset transistor, a source follower transistor and a row selection transistor, wherein the drain of the reset transistor is connected to the drain of the source follower transistor, the source of the reset transistor is connected to the drain of the transfer transistor and the gate of the source follower transistor, the source of the source follower transistor is connected to the source of the row selection transistor, the gate of the row selection transistor is connected to the control end of the row scanner, the drain of the row selection transistor serves as the voltage output end of the photosensitive pixel unit, the source of the transfer transistor is connected to the negative electrode of the photodiode, and the positive electrode of the photodiode is grounded.

[0011] Furthermore, the gate of the reset transistor is used to input a reset signal, the gate of the transmission transistor is used to input a transmission signal, and the drain of the reset transistor and the drain of the source follower transistor are used to input a regulated voltage.

[0012] Furthermore, the bias voltage of the current mirror is adjusted to provide a stable saturation current for each column of the photosensitive pixel units, so that the trusted image sensor is in a digital imaging working mode.

[0013] Furthermore, the bias voltage of the tail current source tube is adjusted so that the source follower transistor in each column of the photosensitive pixel unit is in the subthreshold region, so that the trusted image sensor is in a strong physical unclonable function working mode, providing the hardware trust root required for security authentication of the trusted image sensor.

[0014] Furthermore, the signal readout circuit includes a column-parallel programmable gain amplifier, an analog-to-digital converter and a latch-shift circuit, the input end of the column-parallel programmable gain amplifier is connected to the output end of each column of the photosensitive pixel unit, the output end of the column-parallel programmable gain amplifier is connected to the input end of the analog-to-digital converter, the output end of the analog-to-digital converter is connected to the latch-shift circuit, and the column scanner is connected to the latch-shift circuit.

[0015] Furthermore, the latch-shift circuit comprises a shift register, and the shift register is composed of a plurality of latches connected in series.

[0016] Furthermore, the analog-to-digital converter is a column-parallel 10-bit successive approximation register type analog-to-digital converter.

[0017] The trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF disclosed in the present invention includes a trusted image sensor, a bias adjustment circuit, a row scanner, a column scanner, a signal readout circuit, a control circuit, an analog selector and a comparator, wherein the trusted image sensor includes a plurality of photosensitive pixel units, each of which can be independently photosensitive to output a corresponding photosensitive output voltage; the bias adjustment circuit is connected to each column of the photosensitive pixel units in the trusted image sensor to bias and adjust the magnitude of the input current in each column of the photosensitive pixel units; each row of the photosensitive pixel units The control signal input end of each pixel unit is connected to the control end of the row scanner; the output end of each column of the photosensitive pixel unit is connected to the signal readout circuit and the analog selector, the analog selector is connected to the comparator, and the comparator is used to output the PUF response; the control circuit is connected to the row scanner and the signal readout circuit to provide the row scanner and the signal readout circuit with a first control signal; the column scanner is connected to the analog selector and the signal readout circuit to provide the analog selector with a second control signal and scan the digital imaging signal in the signal readout circuit by column. In the above circuit structure, by multiplexing multiple photosensitive pixel units in the trusted image sensor, through the row scanner, column scanner, bias adjustment circuit, signal readout circuit, analog selector and comparator, not only the digital imaging signal can be output, but also the PUF response can be output, so as to provide the hardware trust root required for security authentication of the trusted image sensor, thereby realizing the trusted sensing function of the image sensor. The trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF disclosed in the present invention solves the problem that the existing image sensors cannot realize the trusted sensing function. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0019] Figure 1 It is an overall architecture diagram of a trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF provided by an embodiment of the present invention;

[0020] Figure 2 It is a shared bias circuit diagram of a digital imaging working mode and a strong physical unclonable function working mode provided by an embodiment of the present invention;

[0021] Figure 3 is a schematic diagram of output voltage under different input excitations provided by an embodiment of the present invention;

[0022] Figure 4 is a schematic diagram of an equivalent circuit of a photosensitive pixel unit provided in an embodiment of the present invention;

[0023] Figure 5 is a schematic diagram of a strong PUF of a trusted image sensor provided by an embodiment of the present invention;

[0024] Figure 6 This is an application effect diagram of a trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF provided by an embodiment of the present invention;

[0025] Figure 7 This is another application effect diagram of a trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF provided by an embodiment of the present invention;

[0026] Figure 8 This is another application effect diagram of the trusted image sensor chip based on the 4T-APS optical reconfigurable strong PUF provided by an embodiment of the present invention;

[0027] Fig. 9 This is another application effect diagram of the trusted image sensor chip based on the 4T-APS optical reconfigurable strong PUF provided by an embodiment of the present invention;

[0028] Fig.10 This is a later application effect diagram of a trusted image sensor chip based on 4T-APS optically reconfigurable strong PUF provided by an embodiment of the present invention;

[0029] Reference numerals:

[0030] PIX, trusted image sensor; 11, bias adjustment circuit; 12, row scanner; 13, column scanner; 14, signal readout circuit; 141, parallel programmable gain amplifier; 142, analog-to-digital converter; 143, latch shift circuit; 15, control circuit; 16, analog selector; 17, comparator. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0033] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0034] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0035] Figures 1 to 10 An embodiment of a trusted image sensor chip based on 4T-APS optically reconfigurable strong PUF provided by the present invention is demonstrated. The trusted image sensor PIX chip based on 4T-APS optically reconfigurable strong PUF of this embodiment includes a trusted image sensor PIX, a bias adjustment circuit 11, a row scanner 12, a column scanner 13, a signal readout circuit 14, a control circuit 15, an analog selector 16 and a comparator 17, wherein the trusted image sensor PIX includes a plurality of photosensitive pixel units, each of which can be independently photosensitive to output a corresponding photosensitive output voltage; the bias adjustment circuit 11 is connected to each column of the photosensitive pixel units in the trusted image sensor PIX to bias-adjust the magnitude of the input current in each column of the photosensitive pixel units; each row of the photosensitive pixel units The control signal input end of each column is connected to the control end of the row scanner 12; the output end of each column of the photosensitive pixel unit is connected to the signal readout circuit 14 and the analog selector 16, and the analog selector 16 is connected to the comparator 17, and the comparator 17 is used to output the PUF response; the control circuit 15 is connected to the row scanner 12 and the signal readout circuit 14 to provide the row scanner 12 and the signal readout circuit 14 with a first control signal; the column scanner 13 is connected to the analog selector 16 and the signal readout circuit 14 to provide the analog selector 16 with a second control signal and scan the digital imaging signal in the signal readout circuit 14 by column. It should be noted that in this embodiment, the first control signal and the second control signal are both control signals with a control function. Figure 1The Row Scanner of each function is a row scanner 12, the Column Scanner is a column scanner 13, the PGA (Programmable Gain Amplifier) ​​is a column-parallel programmable gain amplifier 141, the SAR ADC (Successive-approximation-register ADC) is a column-parallel 10-bit successive approximation register type analog-to-digital converter 142, the Latches are latches, the Shift Registers are shift registers, the Controller is a control circuit 15, the Analog Mux is an analog selector 16, the Comparator is a comparator 17, and the Pixel is a photosensitive pixel unit. It should also be noted that in this embodiment, by multiplexing multiple photosensitive pixel units in the trusted image sensor PIX, through the row scanner 12, the column scanner 13, the bias adjustment circuit 11, the signal readout circuit 14, the analog selector 16 and the comparator 17, not only a digital imaging signal can be output, but also a PUF response can be output, so as to provide the hardware trust root required for security authentication of the trusted image sensor PIX, thereby realizing the trusted sensing function of the image sensor.

[0036] See also Figure 1 In one embodiment, such as the present embodiment, the signal readout circuit 14 includes a column-parallel programmable gain amplifier 141, an analog-to-digital converter 142, and a latch-shift circuit 143. The input end of the column-parallel programmable gain amplifier 141 is connected to the output end of each column of the photosensitive pixel unit, the output end of the column-parallel programmable gain amplifier 141 is connected to the input end of the analog-to-digital converter 142, the output end of the analog-to-digital converter 142 is connected to the latch-shift circuit 143, and the column scanner 13 is connected to the latch-shift circuit 143. Specifically, the latch-shift circuit 143 includes a shift register, and the shift register is composed of a plurality of latches connected in series. More specifically, the analog-to-digital converter 142 is a column-parallel 10-bit successive approximation register type analog-to-digital converter 142. Please refer to Figure 2In one embodiment, such as the present embodiment, the bias adjustment circuit 11 includes a tail current source tube and a current mirror, and the tail current source tube and the current mirror are respectively connected to the voltage output terminal of each photosensitive pixel unit in a column of the photosensitive pixel units. Specifically, the drain of the tail current source tube is connected to the current mirror and the photosensitive pixel unit, its source is grounded, and its gate is used to input a reference voltage. The photosensitive pixel unit includes a photodiode, a transfer transistor, a reset transistor, a source follower transistor, and a row selection transistor, wherein the drain of the reset transistor is connected to the drain of the source follower transistor, the source of the reset transistor is connected to the drain of the transfer transistor and the gate of the source follower transistor, the source of the source follower transistor is connected to the source of the row selection transistor, the gate of the row selection transistor is connected to the control end of the row scanner 12, the drain of the row selection transistor serves as the voltage output terminal of the photosensitive pixel unit, the source of the transfer transistor is connected to the negative electrode of the photodiode, and the positive electrode of the photodiode is grounded. More specifically, the gate of the reset transistor is used to input a reset signal, the gate of the transmission transistor is used to input a transmission signal, and the drain of the reset transistor and the drain of the source follower transistor are used to input a regulated voltage. It should be noted that, in this embodiment, the bias voltage of the current mirror is adjusted to provide a stable saturation current for each column of the photosensitive pixel unit, so that the trusted image sensor PIX is in a digital imaging working mode; the bias voltage of the tail current source tube is adjusted to make the source follower transistor in each column of the photosensitive pixel unit in a subthreshold region, so that the trusted image sensor PIX is in a strong physical unclonable function working mode, providing the hardware trust root required for security authentication of the trusted image sensor PIX. It should be noted that in this embodiment, in the digital imaging working mode, the trusted image sensor PIX first passes through the PGA. The PGA has two main functions: one is to obtain and amplify the output signal from the photosensitive pixel unit to increase the effective input range of the SAR ADC; the other is to completely eliminate the reset noise in the pixel by subtracting the signal voltage from the reset voltage, wherein the reset voltage is the voltage output by the trusted image sensor PIX after the photosensitive pixel is reset; the signal voltage is the voltage output by the trusted image sensor PIX after the pixel is exposed for a period of time. The SAR ADC is responsible for quantizing the amplified voltage signal into a digital signal, and then storing it in the latch in the shift register for reading by the column scanner 13. In the strong physical unclonable function working mode, the trusted image sensor PIX selects pixels in two columns of photosensitive pixel units through the analog selector 16 to enter the comparator 17 to output the PUF response. At this time, both the PGA and the SAR ADC can be turned off to reduce power consumption.

[0037] It should also be noted that, in this embodiment, in the digital imaging working mode, the column lines in the trusted image sensor PIX are biased using a shared constant current source (such as a current mirror) so that the source follower transistor M in the selected photosensitive pixel unit is SF Works in the saturation region. In the strong physical unclonable function working mode, by adding a subthreshold current load M bias , M bias The source follower transistor M in the selected photosensitive pixel unit is called the tail current source tube. SF can be alternately biased into the subthreshold region, so that the output voltage V out and bias current I bias A nonlinear exponential relationship is formed between them. In addition, the row scanner 12 based on the shift register widely used in the digital imaging working mode is reused to provide input excitation for the optically reconfigurable strong PUF. In this way, each column of pixels can have up to 2 M -1 different selection combinations. Providing a uniform and constant reset signal V RST In the case of M -1 combination of output voltage V out And, by multiplexing the column scanner 13 in the digital imaging working mode, two column entropy sources V can be arbitrarily selected. out Compare and digitize, and then output its response. Therefore, the total CRP (Challenge-Response Pairs) space of the strong PUF proposed in this embodiment can be further expanded. As we all know, the image resolution in the digital imaging working mode can reach 12 million (i.e. 3120×4160) or even 50 million (i.e. 6144×8192). Therefore, based on the proposed strong PUF design, its total CRP space can reach 1.41×10 946 and 1.13×10 1857 At the same time, it is worth emphasizing that the above optically reconfigurable strong PUF with huge CRP space can be realized by adding a single-transistor subthreshold current load shared by N columns and a 2-input digital comparator 17 with ultra-low design overhead. In addition, by turning off the increased current load M bias , the proposed optically reconfigurable strong PUF can be disabled without any impact on the digital imaging working mode.

[0038] In one embodiment, such as the present embodiment, after the image sensor is fused with the strong PUF, a theoretical analysis of the anti-machine learning attack of the fused trusted image sensor chip is performed, specifically including: (1) PUF entropy source analysis; (2) anti-machine learning attack analysis.

[0039] (1) PUF entropy source analysis

[0040] As mentioned above, the photosensitive pixel unit of the image sensor is designed with a 4T-APS structure, and the strong PUF in the optical reconfigurable sensor reuses the trusted image sensor as its entropy source. It should be noted that in the trusted image sensor, each photosensitive pixel unit includes a photodiode PD, a transfer transistor M TG , reset transistor M RST , source follower transistor M SF and the row selection transistor M SEL , and is surrounded by a guard ring composed of N-well / deep N-well layers to reduce crosstalk between pixels. In this embodiment, two spatial non-uniformities existing in mainstream CIS pixels can be used as entropy, namely spatial dark signal non-uniformity (Dark Signal Non-Uniformity, DSNU) and photo response non-uniformity (Photo Response Non-Uniformity, PRNU). By utilizing the different physical mechanisms behind these two entropy sources, the DSNU-based PUF generated in a dark environment can be optically reconstructed by irradiating uniform monochromatic light, which is equivalent to white light in actual application scenarios plus the widely used RGB color filter array.

[0041] In the DSNU-based strong physical unclonable function working mode, the reset transistor M RST Normally open, the pass transistor M TG Normally closed, even if there is light, the photocurrent generated by the photodiode PD will not affect the node voltage V FD At this time, the floating diffusion FD node voltage is expressed as formula (1):

[0042] V FD =V DD -V th,RST (1)

[0043] In formula (1), V th,RST Indicates M RST The threshold voltage of M SF The voltage output on the rear line can be expressed as formula (2):

[0044] V out =A v ·V FD =A v ·(V DD -V th,RST ) (2)

[0045] In formula (2), A v It is M SFThe voltage gain of the reset transistor M is RST and source follower transistor M SF That is, due to the inevitable process deviations in the manufacturing process of the trusted image sensor chip, they will inevitably affect the output voltage on the column line. In order to verify the distribution characteristics of the proposed entropy source, the entropy source of the optical reconfigurable strong PUF in the DSNU mode is first statistically analyzed, such as Figure 3 As shown, the output voltage V based on 4096 different input excitations out The statistical histogram of V out The mean and standard deviation are 765 mV and 10.38 mV, respectively.

[0046] When operating in the PRNU-based strong physical unclonable function mode, the reset transistor M in the selected photosensitive pixel unit on the column line RST and the transfer transistor M TG During the response generation period after input stimulation, the photodiode is in a normally open state, and the voltage of the photodiode PD is consistent with the voltage of the FD node. If an external reverse voltage is applied to the photodiode PD, it works in the photoconductive mode, which is like a variable resistor whose resistance changes with the change of light intensity. The photocurrent flows through the external circuit that provides the bias voltage, and the output current or voltage can be measured. Therefore, the equivalent circuit schematic diagram of the photosensitive pixel unit is as follows Figure 4 shown.

[0047] exist Figure 4 In the figure, the photodiode PD can be represented by a current source and an ideal diode in parallel. The ideal diode represents the PN junction and the current source I PH ,I D and I R They represent the photocurrent, dark current and noise current generated by the incident light respectively; the junction capacitance C J And the parallel resistor R SH Connected in parallel with other components. Series resistor R S In series with all components in this model, R L Represents the reset transistor M RST The voltage on the FD node can be expressed as shown in formula (3):

[0048]

[0049] Through the source follower transistor M SF After buffering, the output voltage on the column line can be expressed as shown in formula (4):

[0050]

[0051] In formula (4), Av is a source follower transistor M SF When the PUF works under illumination, the output voltage on the column line is not only proportional to the voltage gain of the reset transistor M RST Threshold voltage and source follower transistor M SF It is related to the voltage gain of the photodiode PD and the incident light. Figure 3 As shown, the output entropy source V based on 4096 different input challenges in the PRNU-based strong physical unclonable function working mode is shown. out The statistical histogram has an average value and standard deviation of 653mV and 15.76mV respectively. The standard deviation is 1.52 times that of the strong physical unclonable function working mode based on DSNU, which effectively improves the stability of PUF.

[0052] (2) Analysis of Anti-machine Learning Attacks

[0053] The most widely used method to resist machine learning modeling attacks is to introduce nonlinear structures to increase the complexity of CRP mapping. Specifically, the exponential relationship between voltage and current when the transistor works in the subthreshold region is used to significantly increase the complexity of the CRP space and, more importantly, reduce the overall power consumption. bias Working in the subthreshold region, its gate voltage V bias Set to 450mV, which can be V out The entropy source introduces a large nonlinearity. Specifically, the IV relationship of a MOS transistor operating in the subthreshold region can be expressed as shown in formula (5):

[0054]

[0055] Figure 5 A schematic diagram of a strong PUF of a trusted image sensor provided by an embodiment of the present invention, wherein a row scanner and a column scanner can provide (M+N)-bit excitation, two columns can be arbitrarily selected from the N columns, and any X pixels can be selected from the M pixels in each column as entropy sources. bias The total current I bias It can be calculated as shown in formula (6):

[0056]

[0057] In formula (6), V bias 、V th,b 、V ds,b , W b , L b 、m b They are the tail current source tube M biasThe gate voltage, threshold voltage, drain-source voltage, effective channel width / length and subthreshold slope coefficient, From formula (6), we can see that considering the inevitable mismatch of process-related parameters in the manufacturing process of all selected pixels, V out The final output value of the input stimulus is a nonlinear and highly complex function.

[0058] For the strong PUF proposed in this embodiment, a mathematical model analysis is performed for any two columns with M photosensitive pixel units in each column. Definition is the 1×M feature vector representing the input stimulus C; is the 1×M row vector of the photosensitive pixel unit affected by process variation, which is modeled as an equivalent conductance. Under the same power supply voltage condition, the selected pixel units sharing the same column line can be regarded as a whole. Therefore, the output response R of the selected two columns of entropy sources after digitization by the comparator can be equivalently expressed as formula (7):

[0059]

[0060] According to formula (7), the final response R is determined by the equivalent conductance value of the selected photosensitive pixel unit. Taking the i-th selected photosensitive pixel unit as an example, according to formula (5), I dsi Yes V ds (i.e. V DD -V out ), then M SF The conductivity value ρ i It can be expressed as formula (8):

[0061]

[0062] From formula (8), we can know that in each selected photosensitive pixel unit M SF The conductivity ρ i Yes V out is a nonlinear function, and V out is jointly affected by all other selected photosensitive pixel units. DD When constant, V out Any change not only affects V ds , which also affects V gsi (i.e. V PDi -V out ), which in turn leads to the i Therefore, each selected pixel M SF The conductivity increases with V ds Dynamically changes, while the output voltage V outIt also changes dynamically with different selected photosensitive pixel units. As a result, the overall parameters required for modeling attacks are greatly increased. Specifically, for M photosensitive pixel units in a column, the conductance of any selected photosensitive pixel unit will be affected by the other (M-1) pixel units. Theoretically, the maximum modeling parameter is M×(2 M-1 +1). In this embodiment, the row photosensitive pixels M=128, the column photosensitive pixels N=128, and the total parameters that the machine learning algorithm needs to model are approximately 2.18×10 40 , this high complexity of model parameters can significantly enhance the robustness of the proposed strong physical unclonable function circuit and reduce the accuracy of accurate modeling using mainstream machine learning algorithms. Even with advanced artificial intelligence algorithms, it is difficult to crack models with the above-mentioned scale parameters within a reasonable time. Therefore, the strong physical unclonable function circuit has good anti-attack properties against advanced artificial intelligence algorithms from the perspective of mathematical models.

[0063] In one embodiment, such as this embodiment, in order to verify the proposed trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF, a standard 65nm CMOS process is used for design and manufacturing in this embodiment, where M and N are both designed to be 128. After measurement, the entire chip area is 2.33×10 6 μm 2 (i.e. 5.52×10 8 F 2 ), dedicated to the proposed strong PUF (i.e., the increased tail current source tube M bias The additional silicon area (including digital comparator) is 6174μm 2 (i.e. 1.46×10 6 F 2 ). This means that more than 99.735% of the area of ​​the entire chip can be reused for the proposed strong PUF, which only adds an ultra-small design overhead of 0.265%. In addition, an FPGA development board (Xilinx DaVinci) is used to provide 256 bits (i.e., 128+128=256) of input excitation for the proposed strong PUF by fully reusing the row scanner and column scanner of the designed trusted image sensor. An ultra-low power of 14.15μW is achieved at a maximum throughput of 2.5Mb / s, and the energy efficiency of the proposed strong PUF is 5.66pJ / bit.

[0064] 1) Reliability

[0065] Reliability is an important quality factor that measures the repeatability of the CRP of a PUF under various environmental factors, such as time-domain noise, power supply voltage, and temperature variations. First, the effect of time-domain noise on its stability is evaluated by repeatedly reading the output response of the PUF at a fixed voltage and temperature. Specifically, under normal voltage-temperature (VT) conditions (25°C, 1.2V), 2000 different stimuli are input to 20 test chips, and the cycle is repeated 5000 times, and then the bit error rate (BER) of the output response is counted. Figure 6 As shown in (a), under 20 test chips, the average worst raw bit error rates of PUF based on DSNU and PRNU are 2.26% and 1.54% respectively; while the average worst raw bit error rates (Unstable Bit Rate, UBR) are 9.21% and 4.84% respectively. Figure 6 As shown in (b), changes in power supply voltage and temperature will further reduce the reliability of PUF, and its bit error rate is evaluated by comparing it with the output response under normal VT working conditions. The bit error rate of 20 test chips under the temperature range of -20℃~100℃ and the power supply voltage range of 1.08V~1.32V is tested under various conditions and the average value is taken. Figure 6 (c) is the two-dimensional distribution of the average bit error rate of the PUF based on DSNU within the above VT variation range. The results show that the worst bit error rate without any correction occurs under the VT condition of (100℃, 1.08V), and the value is 13.3%; while the average bit error rate of the PUF based on PRNU, the worst bit error rate without any correction occurs under the VT condition of (100℃, 1.08V), and the value is 8.4%. Figure 6 (d) as shown.

[0066] 2) Randomness

[0067] The randomness of the strong PUF proposed in this embodiment can be evaluated by the autocorrelation function (ACF) and the NIST random test. The randomness well reflects the unpredictability of the strong PUF to unknown responses. As we all know, ACF is an important indicator to measure the randomness of a bit stream. For a bit stream generated by white noise, its ACF value is close to zero at any lag value except zero. For a strong PUF with an exponential CRP space in a limited silicon area, the same PUF bit unit shared by different CPRs may cause correlation in the output response. Here, ACF is first used to quantify the above correlation, such as Figure 7As shown in the figure, under the test of 200,000 stimulus generation responses, the ACF values ​​of the PUF based on DSNU and PRNU are equal to ±0.01369 and ±0.01104 respectively at a confidence level of 95%, indicating strong randomness. In addition, 10 million responses were collected for the proposed strong PUF and divided into 20 groups, and the randomness was further statistically evaluated using the NISTPub.800-22 test suite. According to the statistical analysis results, it is shown that the PUF based on DSNU and PRNU have passed all relevant NIST tests, and their corresponding P-Val are greater than 0.01.

[0068] 3) Uniqueness

[0069] Uniqueness is used to quantitatively describe the distinguishability between different PUF chips, which is also called the Hamming distance between PUFs. Applying 2000 identical stimuli at reference temperature and voltage (25°C, 1.2V), a total of 20 test chips can be calculated. The Hamming distance between PUFs is normalized and plotted on Figure 8 The uniqueness is defined as the average value of the best fitting Gaussian curve of the above inter-PUF Hamming distance distribution. The results show that the inter-chip Hamming distances of PUFs based on DSNU and PRNU are 0.4969 and 0.5013, respectively, both close to the ideal value of 0.5. In addition, this experiment also measured and calculated the intra-PUF Hamming distance to evaluate its robustness. In order to evaluate the robustness of PUF to time domain noise, the 2000-bit output response of the PUF is read repeatedly 500 times, and the Hamming distance between any two of the 500 reads of the 2000-bit output response is calculated. Figure 8 As shown, the intra-chip Hamming distance distribution of the DSNU-based PUF is μ=0.0097 and σ=0.0012, and the separation degree from the inter-chip Hamming distance distribution is 51.2 times; the intra-chip Hamming distance distribution of the PRNU-based PUF is μ=0.0053 and σ=0.0016, and the separation degree from the inter-chip Hamming distance distribution is 94.6 times.

[0070] 4) Optical reconfigurability

[0071] Reconfigurability is a basic property of reconfigurable PUF and can also be evaluated by the Hamming distance between chips. This embodiment can achieve optically reconfigurable CRP by changing the state of external ambient light, that is, by switching from DSNU mode to PRNU mode. In the experiment, the CRP was reconfigured in a dark environment and with a power of 0.545mW / cm 2Under the 550nm light source, 50 5120-bit PUF instances were extracted from 10 test chips. The results show that the distribution mean of the 50 reconfiguration Hamming distances is 0.4896 and the standard deviation is 0.0466. The Pearson correlation coefficient (PCC) can show the correlation between two sequences. If each variable has N scalar observations, the PCC definition is as shown in formula (9):

[0072]

[0073] In formula (9), μ A and σ A are the mean and standard deviation of A, μ B and σ B are the mean and standard deviation of B. To further analyze the correlation between different reconstruction responses, PCC is calculated. Based on 50 5120-bit PUF instances extracted from 10 test chips, the reported average PCC is 0.1866, verifying that the PUF bitstreams obtained from DSNU and PRNU are independent of each other, as shown in Fig. 9 The results show that the reconfigurable PUF successfully achieves the generation of uncorrelated CRP between DSNU and PRNU patterns.

[0074] 5) Anti-machine learning attack performance

[0075] Modeling attacks are tools used by attackers to try to break the authentication provided by strong PUFs, so the ability of strong PUFs to resist modeling attacks is considered the ultimate measure of their security. These attacks rely on the power of machine learning algorithms to accurately predict unknown CRPs by collecting a small number of CRPs to model the internal parameters of strong PUFs. For most existing strong PUFs, such as APUFs, their corresponding model functions can be represented by a linear superposition of internal delays, making them vulnerable to machine learning modeling attacks. The nonlinear strong PUF proposed in this embodiment exhibits a large number of unknown parameters, which increases the complexity of machine learning modeling attacks.

[0076] In order to evaluate the ability to resist machine learning attacks, we used four mainstream machine learning algorithms to attack the proposed strong PUF, including ANN based on 80×80×80 multi-layer perception, LR using gradient descent optimizer, SVM using radial basis function (RBF), and CMA-ES. In the experiment, 40 million CRPs were randomly collected from a prototype chip as a training set, and 5 million CRPs were collected as a test set. The number of CRPs in the training set was changed from 10 million to 40 million, and ANN, LR, SVM and CMA-ES algorithms were used for training and prediction. Fig.10 The relationship between the prediction accuracy of various machine learning algorithms and the number of CRPs used to train these algorithms. It can be seen that the prediction accuracy of PUF based on DSNU and PRNU for all the above machine learning algorithms can be well maintained at around 50%, which is equivalent to random prediction.

[0077] 6) Performance comparison

[0078] The performance of the trusted image sensor chip in this embodiment is compared with that of the most advanced ASIC strong PUF (in this embodiment, there are 4 comparisons, among which comparison 1 is the performance result disclosed in "Zhuang H, Xi X, Sun N, et al. A strong subthreshold current array PUF resilient to machine learning attacks [J]. IEEE Transactions on Circuits and Systems I: Regular Papers, 2019, 67 (1): 135-144."; comparison 2 is the performance result disclosed in "Venkatesh A, Venkatasubramaniyan A B, Xi X, et al. 0.3 pJ / bit machine learning resistant strong PUF using subthreshold voltage divider array [J]. IEEE Transactions on Circuits and Systems II: Express Briefs, 2019, 67 (8): 1394-1398."; comparison 3 is the performance result disclosed in "Liu J, Zhu Y, Chan CH, et al. A 0.04% BER strong PUF with cell-bias-based CRPs filtering and background The performance results disclosed in “offsetcalibration[J].IEEE Transactions on Circuits and Systems I:RegularPapers,2020,67(11):3853-3865.” are compared with the performance results disclosed in “Zhang J,Xu C,Law MK,et al.A4T / Cell Amplifier-Chain-Based XOR PUF With Strong Machine Learning AttackResilience[J].IEEE Transactions on Circuits and Systems I:RegularPapers,2021,69(1):366-377.”). The strong PUF proposed in this embodiment has an ultra-high hardware reuse rate of 99.735%, and the silicon area overhead dedicated to designing PUF is as compact as only 1.46×10 6 F 2Even when the pixel array in the digital imaging mode is only 128×128, the CRP space of the strong PUF is 2.77×10 42 , and the CRP space can be further expanded by designing a larger digital imaging working mode pixel array. The CRP space is normalized to 5.52×10 8 F 2 The calculated area efficiency is 5.01×10 33 bit / F 2 , significantly exceeding the area efficiency of other ASIC strong PUFs by 19 to 22 orders of magnitude.

[0079] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF, characterized in that: The invention comprises a trusted image sensor, a bias adjustment circuit, a row scanner, a column scanner, a signal readout circuit, a control circuit, an analog selector and a comparator, wherein: The trusted image sensor includes a plurality of photosensitive pixel units, each of which can independently sense light to output a corresponding light-sensitive output voltage; The bias adjustment circuit is connected to each column of the photosensitive pixel units in the trusted image sensor to perform bias adjustment on the magnitude of the input current in each column of the photosensitive pixel units; The control signal input end of each row of the photosensitive pixel units is connected to the control end of the row scanner; The output end of each column of the photosensitive pixel unit is connected to the signal readout circuit and the analog selector, the analog selector is connected to the comparator, and the comparator is used to output a PUF response; The control circuit is connected to the row scanner and the signal readout circuit to provide a first control signal to the row scanner and the signal readout circuit; The column scanner is connected to the analog selector and the signal readout circuit to provide a second control signal to the analog selector and scan the digital imaging signal in the signal readout circuit by column; The bias adjustment circuit includes a tail current source tube and a current mirror, and the tail current source tube and the current mirror are respectively connected to the voltage output end of each photosensitive pixel unit in a column of photosensitive pixel units; The signal readout circuit includes a column-parallel programmable gain amplifier, an analog-to-digital converter and a latch-shift circuit. The input end of the column-parallel programmable gain amplifier is connected to the output end of each column of the photosensitive pixel unit, the output end of the column-parallel programmable gain amplifier is connected to the input end of the analog-to-digital converter, the output end of the analog-to-digital converter is connected to the latch-shift circuit, and the column scanner is connected to the latch-shift circuit.

2. The trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF according to claim 1 is characterized in that: The drain of the tail current source tube is connected to the current mirror and the photosensitive pixel unit, the source thereof is grounded, and the gate thereof is used for inputting a reference voltage.

3. The trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF according to claim 2 is characterized in that: The photosensitive pixel unit includes a photodiode, a transfer transistor, a reset transistor, a source follower transistor and a row selection transistor, wherein: The drain of the reset transistor is connected to the drain of the source follower transistor, the source of the reset transistor is connected to the drain of the transfer transistor and the gate of the source follower transistor, the source of the source follower transistor is connected to the source of the row selection transistor, the gate of the row selection transistor is connected to the control end of the row scanner, the drain of the row selection transistor serves as the voltage output end of the photosensitive pixel unit, the source of the transfer transistor is connected to the cathode of the photodiode, and the anode of the photodiode is grounded.

4. The trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF according to claim 3 is characterized in that: The gate of the reset transistor is used to input a reset signal, the gate of the transmission crystal is used to input a transmission signal, and the drain of the reset transistor and the drain of the source follower transistor are used to input a regulated voltage.

5. The trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF according to claim 1 is characterized in that: The bias voltage of the current mirror is adjusted to provide a stable saturation current for each column of the photosensitive pixel units, so that the trusted image sensor is in a digital imaging working mode.

6. The trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF according to claim 1 is characterized in that: The bias voltage of the tail current source tube is adjusted so that the source follower transistor in each column of the photosensitive pixel unit is in the subthreshold region, so that the trusted image sensor is in a strong physical unclonable function working mode, providing the hardware trust root required for security authentication of the trusted image sensor.

7. The trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF according to claim 1 is characterized in that: The latch-shift circuit comprises a shift register, and the shift register is composed of a plurality of latches connected in series.

8. The trusted image sensor chip based on 4T-APS optical reconfigurable strong PUF according to claim 1 is characterized in that: The analog-to-digital converter is a column-parallel 10-bit successive approximation register type analog-to-digital converter.

Images