Implementation method and implementation device of physical unclonable function
By randomly setting weight parameters in a neural network structure and using an anti-variable memory array, a physical non-clone function that efficiently resists modeling attacks in hardware security is realized, solving the problems of large overhead and insufficient anti-attack capabilities in existing PUF hardware.
Patent Information
- Application Number
- CN202010643818.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-07-07
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2040-07-07
AI Technical Summary
Existing physical non-cloneable functions (PUFs) have problems with large hardware overhead and insufficient attack resistance when facing hardware reuse and modeling attacks, especially simple hardware reuse solutions cannot effectively resist various modeling attacks.
Using a physical non-clone function based on a neural network structure, the weight parameter matrix is randomly set in the neural network structure, and the resistance-variable memory array is used to realize the randomness of the weight parameters, and the key value is generated, combining the fully connected structure and activation function. Circuit to improve the number of excitation response pairs and data complexity.
Achieve huge amount of stimulus response pairs on smaller hardware overhead, and improve the PUF's anti-modeling attack capability, enhancing the randomness and attack resistance of key values through complex connection relationships of neural network-like structures.
Smart Images

Figure CN111797438B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present disclosure relate to a method and device for implementing a physical unclonable function. Background Art
[0002] With the rapid development of information technology, hardware security faces many threats (e.g., hardware Trojans, intellectual property theft, integrated circuit reverse engineering, side channel attacks, etc.). Among the many solutions, physical unclonable functions are considered to be a feasible technical solution for hardware security protection (e.g., key protection, hardware authentication) due to their inherent randomness, reproducibility, and compatibility with micro-nano processing technology. Summary of the invention
[0003] At least one embodiment of the present disclosure provides a method for implementing a physical unclonable function based on a neural network-like structure, wherein the neural network-like structure includes multiple neuron layer circuits and a weight parameter matrix between the multiple neuron layer circuits, and the method includes: performing a setting operation on the neural network-like structure so that at least part of the weight parameters in the weight parameter matrix are randomly set; providing an input electrical signal to the neural network-like structure, and the neural network-like structure provides an output electrical signal based on the input electrical signal and multiple weight parameters of the weight parameter matrix; and determining a key value of the physical unclonable function based on the output electrical signal.
[0004] For example, the implementation method provided by at least one embodiment of the present disclosure also includes: using a resistive memory array to implement a weight parameter matrix of the neural network-like structure, the conductance values of multiple resistive memories in the resistive memory array respectively correspond to multiple weight parameters of the weight parameter matrix, and each resistive memory in the resistive memory array includes a first end connected to a previous neuron layer circuit and a second end connected to a subsequent neuron layer circuit.
[0005] For example, in the implementation method provided by at least one embodiment of the present disclosure, a fully connected structure is adopted between the multiple neuron layer circuits.
[0006] For example, in the implementation method provided by at least one embodiment of the present disclosure, an activation function circuit is also included between the multiple neuron layer circuits.
[0007] For example, in the implementation method provided by at least one embodiment of the present disclosure, a setting operation is performed on the neural network-like structure so that at least part of the weight parameters in the weight parameter matrix are randomly set, including: performing at least one reset operation on at least part of the resistive memory in the resistive memory array so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set.
[0008] For example, in the implementation method provided by at least one embodiment of the present disclosure, a setting operation is performed on the neural network-like structure so that at least part of the weight parameters in the weight parameter matrix are randomly set, including: performing at least one set operation on at least part of the resistive memory in the resistive memory array so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set.
[0009] For example, in the implementation method provided by at least one embodiment of the present disclosure, a setting operation is performed on the neural network-like structure so that at least part of the weight parameters in the weight parameter matrix are randomly set, including: performing a combination of at least one reset operation and at least one set operation on at least part of the resistive memory in the resistive memory array so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set.
[0010] For example, in the implementation method provided by at least one embodiment of the present disclosure, the reset operation reduces the conductance value of the operated resistive memory, and the set operation increases the conductance value of the operated resistive memory.
[0011] For example, in the implementation method provided by at least one embodiment of the present disclosure, an input electrical signal is provided to the neural network-like structure, and the neural network-like structure provides an output electrical signal based on the input electrical signal and multiple weight parameters of the weight parameter matrix, including: applying a voltage to at least part of the resistive memory in the resistive memory array to output a current value, the voltage being the input electrical signal, and the current value being the output electrical signal.
[0012] For example, in the implementation method provided by at least one embodiment of the present disclosure, determining the key value of the physical unclonable function based on the output electrical signal includes: comparing the output current value with a reference current value, and outputting the key value of the physical unclonable function based on the comparison result.
[0013] At least one embodiment of the present disclosure also provides an implementation device of a physical unclonable function based on a neural network-like structure, the implementation device of the physical unclonable function comprising: a setting circuit, an input circuit, a storage and calculation circuit, and an output circuit, the storage and calculation circuit comprising the neural network-like structure, the neural network-like structure comprising a plurality of neuron layer circuits and a weight parameter matrix between the plurality of neuron layer circuits, the setting circuit being configured to perform a setting operation on the storage and calculation circuit so that at least part of the weight parameters in the weight parameter matrix are randomly set; the input circuit being configured to provide an input electrical signal to the storage and calculation circuit; the storage and calculation circuit being configured to provide an output electrical signal based on the input electrical signal and a plurality of weight parameters of the weight parameter matrix; the output circuit being configured to determine a key value of the physical unclonable function based on the output electrical signal.
[0014] For example, in the implementation device provided by at least one embodiment of the present disclosure, the storage and computing circuit includes a resistive memory array, the resistive memory array serves as a weight parameter matrix of the neural network-like structure, the conductance values of multiple resistive memories in the resistive memory array respectively correspond to multiple weight parameters of the weight parameter matrix, and each resistive memory in the resistive memory array includes a first end connected to a previous neuron layer circuit and a second end connected to a next neuron layer circuit.
[0015] For example, in the implementation device provided by at least one embodiment of the present disclosure, the first end of the first part of the resistive memory of the resistive memory array is connected to the input circuit to receive the input electrical signal, and the second end of the second part of the resistive memory of the resistive memory array is connected to the output circuit to output the output electrical signal.
[0016] For example, in the implementation device provided by at least one embodiment of the present disclosure, the setting circuit is configured to: perform at least one reset operation on at least part of the resistive memory in the resistive memory array so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set.
[0017] For example, in the implementation device provided by at least one embodiment of the present disclosure, the setting circuit is configured to: perform at least one set operation on at least part of the resistive memory in the resistive memory array so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set.
[0018] For example, in the implementation device provided by at least one embodiment of the present disclosure, the setting circuit is configured to: perform a combination of at least one reset operation and at least one set operation on at least part of the resistive memory in the resistive memory array, so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set.
[0019] For example, in the implementation device provided by at least one embodiment of the present disclosure, the input circuit is configured to apply a voltage to at least part of the resistive memory in the resistive memory array to output a current, and the storage calculation circuit is configured to obtain and output a current value of the current, the voltage is the input electrical signal, and the current value is the output electrical signal.
[0020] For example, in the implementation device provided by at least one embodiment of the present disclosure, the output circuit is configured to: compare the output current value with a reference current value, and output the key value of the physical unclonable function based on the comparison result. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments will be briefly introduced below. Obviously, the drawings in the following description only relate to some embodiments of the present disclosure, but are not intended to limit the present disclosure.
[0022] Figure 1 A flowchart of a method for implementing a physical unclonable function based on a neural network-like structure provided for at least one embodiment of the present disclosure;
[0023] Figure 2 A schematic diagram illustrating an example of a neural network-like structure according to an embodiment of the present disclosure;
[0024] Figure 3A A schematic diagram of a neural network-like structure including four nodes according to an embodiment of the present disclosure is illustrated;
[0025] Figure 3B Diagram showing the implementation Figure 3A A schematic diagram of a memory array of a weight parameter matrix of a neural network-like structure shown;
[0026] Figure 4 A schematic diagram illustrating the electrical characteristics of a resistive memory is shown;
[0027] Figure 5 The figure shows the current distribution diagram of multiple resistive memory;
[0028] Figure 6 A distribution diagram of key values generated by a physical unclonable function according to an embodiment of the present disclosure is illustrated;
[0029] Figure 7 A schematic block diagram of a device for implementing a physical unclonable function based on a neural network-like structure provided for at least one embodiment of the present disclosure. DETAILED DESCRIPTION
[0030] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings of the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.
[0031] Unless otherwise defined, the technical terms or scientific terms used in the present disclosure should be understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, similar words such as "one", "one" or "the" do not indicate quantity restrictions, but indicate that there is at least one. Similar words such as "include" or "comprise" mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Similar words such as "connect" or "connected" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0032] For the application of Physical Unclonable Function (PUF) for hardware authentication, a large number of stimulus-response pairs are required to cope with possible replay attacks. At present, PUF mostly adopts a simple hardware reuse scheme to achieve a large number of stimulus-response pairs. For example, the arbitrator PUF based on CMOS technology uses the delay randomness of the circuit as the random source, and expands the number of stimulus-response pairs by reusing the delay unit. For example, the PUF based on resistive memory uses the resistance randomness in the resistance conversion process as the random source, and generates a key by randomly comparing the resistance values of two resistive memories in the resistive memory array to achieve a large number of stimulus-response pairs. For example, for an M×N resistive memory array, up to However, the above simple hardware reuse scheme has the disadvantages of large hardware overhead and inability to effectively resist various modeling attacks due to the low data complexity of the stimulus-response pairs.
[0033] It should be noted that the stimulus refers to the input and the response refers to the output. For example, in the present disclosure, the stimulus may specifically refer to the input of the PUF, and the response may specifically refer to the key value output by the PUF. However, since the stimulus-response pair is intended to describe the corresponding relationship between the input and the output, the present disclosure does not limit this. For example, the response may also specifically refer to the intermediate number used to generate the PUF key value.
[0034] At least one embodiment of the present disclosure provides a method for implementing a physical unclonable function based on a neural network-like structure, wherein the neural network-like structure includes multiple neuron layer circuits and a weight parameter matrix between the multiple neuron layer circuits, and the method includes: performing a setting operation on the neural network-like structure so that at least part of the weight parameters in the weight parameter matrix are randomly set; providing an input electrical signal to the neural network-like structure, and the neural network-like structure provides an output electrical signal based on the input electrical signal and multiple weight parameters of the weight parameter matrix; and determining a key value of the physical unclonable function based on the output electrical signal.
[0035] At least one embodiment of the present disclosure also provides an implementation device corresponding to the implementation method of the above-mentioned physical unclonable function.
[0036] The implementation method and device of the above-mentioned physical unclonable function provided by the embodiments of the present disclosure can realize a huge number of stimulus-response pairs with a relatively small hardware overhead by utilizing a neural network-like structure, and can also increase the data complexity of the stimulus-response pairs by utilizing the complex connection relationship within the neural network-like structure, thereby improving the anti-modeling attack capability of the implemented PUF.
[0037] The embodiments and examples of the present disclosure are described in detail below with reference to the accompanying drawings.
[0038] Figure 1 A flowchart of a method for implementing a physical unclonable function based on a neural network-like structure is provided for at least one embodiment of the present disclosure. For example, the neural network-like structure includes a plurality of neuron layer circuits and a weight parameter matrix between the plurality of neuron layer circuits. Figure 1 As shown, the implementation method includes steps S110, S120 and S130.
[0039] Step S110: performing a setting operation on the neural network-like structure so that at least part of the weight parameters in the weight parameter matrix are randomly set.
[0040] Step S120: providing an input electrical signal to the neural network-like structure, and the neural network-like structure provides an output electrical signal based on the input electrical signal and multiple weight parameters of the weight parameter matrix.
[0041] Step S130: Determine the key value of the physical unclonable function based on the output electrical signal.
[0042] For example, Figure 2 FIG. 1 is a schematic diagram showing an example of a neural network structure according to an embodiment of the present disclosure. Figure 2 As shown, the neural network structure includes a plurality of neuron layer circuits (e.g., neuron layer circuits 10 and 20) and a weight parameter matrix (e.g., weight parameter matrix 30) between the plurality of neuron layer circuits, wherein each neuron layer circuit includes a plurality of nodes (e.g., Figure 2 The multiple neuron layer circuits adopt a fully connected structure (that is, each node in the neuron layer circuit is connected to all the nodes of the previous neuron layer circuit).
[0043] It should be noted that Figure 2 The neural network structure shown is exemplary and not intended to limit the present disclosure. For example, a non-fully connected structure may be used between multiple neuron layer circuits. For example, an activation function circuit ( Figure 2 (not shown in the figure). For example, the activation function circuit may include a sigmoid function circuit, a tanh function circuit, or a relu function circuit.
[0044] It should be noted that, in the embodiments of the present disclosure, the neuron layer circuit may or may not have a corresponding circuit structure. For example, the first layer of neuron layer circuits may be an interface circuit to receive input electrical signals. For example, the intermediate layers of neuron layer circuits may be interfaces (for example, directly connected by wires). For example, the intermediate layers of neuron layer circuits may be or include current-to-voltage circuits, analog-to-digital conversion circuits, or storage circuits.
[0045] In the embodiments of the present disclosure, although the neural network-like structure has a network structure similar to the neural network structure, there are also differences between the neural network-like structure and the neural network structure. For example, in a conventional neural network structure, a training operation (e.g., an on-chip training operation, an off-chip training operation, or a combination of the two) is required for the weight parameter matrix so that each weight parameter in the weight parameter matrix (i.e., each weight parameter of the weight parameter matrix) is a specific value rather than a random value; in a neural network-like structure, there is no need to perform a training operation on the weight parameter matrix, and at least part of the weight parameters in the weight parameter matrix are randomly set to random values.
[0046] For example, in an embodiment of the present disclosure, a setting operation is performed on a neural network-like structure so that at least part of the weight parameters in the weight parameter matrix are randomly set. It should be noted that, generally, the higher the proportion of randomly set weight parameters in the weight parameter matrix, the better the randomness of the key value generated by the final physical unclonable function. For example, in some examples, a setting operation is performed on a neural network-like structure so that all weight parameters in the weight parameter matrix are randomly set. It should also be noted that the setting operation may be based on the specific implementation form of the neural network-like structure. For example, when the neural network-like structure is implemented by a resistive memory array (described in the following embodiments), the setting operation may reset all resistive memories in the resistive memory array. The specific introduction will be described below, and the embodiments of the present disclosure are not limited to this.
[0047] For example, refer to Figure 2 , an input electrical signal can be provided to the first neuron layer circuit of the neural network-like structure, and the neural network-like structure provides an output electrical signal in the last neuron layer circuit based on the input electrical signal and multiple weight parameters of the weight parameter matrix. For example, the input electrical signal can be a set of voltage signals. For example, the input electrical signal can also be a set of numerical values, which can be converted into a form suitable for the neural network-like structure through digital-to-analog conversion. For example, the output electrical signal can be a set of current signals. For example, the output electrical signal can also be a set of numerical values obtained through analog-to-digital conversion. It should be noted that the form of the input electrical signal and the output electrical signal can be determined according to the specific implementation form of the neural network-like structure, and the embodiments of the present disclosure are not limited to this.
[0048] For example, the output electrical signal may be compared with the reference signal, and the key value of the physical unclonable function may be output based on the comparison result. For example, the output electrical signal may be analog-to-digital converted, and the key value of the physical unclonable function may be output based on the analog-to-digital conversion result. It should be noted that the specific method for determining the key value of the physical unclonable function may be determined according to the specific situation, and the embodiments of the present disclosure are not limited to this.
[0049] In the embodiments of the present disclosure, the input electrical signal can be regarded as an excitation, and the output electrical signal can be regarded as a response. Alternatively, the input electrical signal can be regarded as an excitation, and the key value of the physical unclonable function can be regarded as a response. The use of the above-mentioned neural network-like structure can greatly increase the number of stimulus-response pairs, and due to the complex connection relationship within the neural network-like structure, the stimulus needs to undergo complex calculations to obtain the corresponding response, thereby increasing the data complexity of the stimulus-response pair, and further improving the ability to resist modeling attacks. In addition, due to the flexible organizational structure of the neural network-like structure, the physical unclonable function can be flexibly adjusted according to the number of key values to be generated.
[0050] For example, in an embodiment of the present disclosure, a neural network-like structure may be implemented by means of software, hardware, firmware, or any combination thereof, and the embodiment of the present disclosure is not limited thereto.
[0051] For instance, in some examples, a resistive memory array is used to implement a weight parameter matrix of a neural network-like structure, the conductance values of multiple resistive memories in the resistive memory array respectively correspond to multiple weight parameters of the weight parameter matrix, and each resistive memory in the resistive memory array includes a first end (e.g., an input end) connected to a previous neuron layer circuit and a second end (e.g., an output end) connected to a subsequent neuron layer circuit.
[0052] For example, the resistive memory may adopt a 1R structure, that is, only one resistive switching element. For example, the resistive memory may also adopt a 1T1R structure, that is, one transistor and one resistive switching element. Of course, other resistive switching memories may also be adopted, and the embodiments of the present disclosure are not limited thereto.
[0053] For example, refer to Figure 3A and Figure 3B , the specific implementation method of the weight parameter matrix of a neural network-like structure using a resistive memory array is explained. Figure 3A A schematic diagram of a neural network-like structure including four nodes according to an embodiment of the present disclosure is illustrated. Figure 3B Diagram showing the implementation Figure 3A Schematic diagram of the memory array of the weight parameter matrix of the neural network-like structure shown.
[0054] like Figure 3A As shown, the neural network structure is 2×2 dimensional, including two neuron layer circuits, each neuron layer circuit includes two nodes, the input electrical signals of the two nodes in the first neuron layer circuit are x1 and x2, and the output electrical signals of the two nodes in the second neuron layer circuit are y1 and y2. The first neuron layer circuit and the second neuron layer circuit adopt a fully connected structure, and the weight parameters include w 11 、w 12 、w 21 and w 22 , the input electrical signal and the output electrical signal satisfy the following relationship: y1 = x1*w 11 +x2*w 21 , y2=x1*w 12 +x2*w 22 .
[0055] like Figure 3B As shown, the resistive memory array is 2×2 dimensional and includes four resistive memory devices in a cross array structure, and the conductance values are g 11 , g 12 , g21 and g 22 , the four resistive memory devices all adopt a 1R structure, one electrode of each resistive memory device is connected to a word line terminal, and the other electrode of each resistive memory device is connected to a bit line terminal. It should be noted that in the embodiments of the present disclosure, the two electrodes of the resistive memory device adopting a 1R structure are described as one electrode and the other electrode respectively, which is only for the purpose of distinction and is not a limitation of the present disclosure. Assume that Figure 3B The bit line terminal in the resistive memory array shown is grounded, and voltages with amplitudes of V1 and V2 are applied to the word line terminals respectively. Then, the currents I1 and I2 on the bit line terminals can be calculated by the conductance value of the resistive memory. The following relationship is satisfied between V1, V2 and I1, I2: I1=V1*g 11 +V2*g 21 , I2=V1*g 12 +V2*g 22 .
[0056] By comparison, the electrical characteristics of the resistive memory crossbar array are exactly the same as the computational relationship between nodes in the neural network structure, namely:
[0057] y1→I1,y2→I2,x1→V1,x2→V2,
[0058] w 11 → 11 ,w 12 → 12 ,w 21 → 21 ,w 22 → 22
[0059] The symbol “→” indicates corresponding to or equivalent to. Therefore, the computational relationship between nodes in the neural network-like structure (i.e., the weight parameter matrix) can be realized by the resistive memory array. For example, the conductance value of each resistive memory in the resistive memory array is used as the weight parameter in the weight parameter matrix, the word line end of each resistive memory is connected to the previous neuron layer circuit as the first end (e.g., the input end), and the bit line end of each resistive memory is connected to the next neuron layer circuit as the second end (e.g., the output end).
[0060] For example, Figure 3BThe resistive memory in the example shown may also adopt a 1T1R structure, wherein the transistor in the resistive memory includes a gate, a source and a drain (the source and drain of the transistor are usually symmetrical in physical structure and can be interchanged). For ease of explanation, it is assumed that the transistor in the resistive memory with a 1T1R structure adopts an N-type transistor, the gate is connected to the word line end, the source is connected to the source line end, and the drain is connected to the bit line end. In the example where the resistive memory adopts a 1T1R structure, the transistor is turned on when a high level is input at the word line end, the source line end of each resistive memory is connected to the previous neuron layer circuit as the first end, and the bit line end of each resistive memory is connected to the next neuron layer circuit as the second end, and the calculation relationship between the nodes in the neural network structure can also be realized. The specific process is similar to that of the 1R structure, which will not be repeated here. It should be noted that the following embodiments also adopt the above assumptions if they involve a resistive memory with a 1T1R structure, but the above assumptions are only exemplary and not a limitation of the present disclosure. For example, the transistor in the resistive memory with a 1T1R structure may also adopt a P-type transistor.
[0061] It should also be noted that the Figure 3A and Figure 3B The examples shown are described, which are exemplary and not limiting of the present disclosure. For example, the computational relationship between the nodes of quasi-neural network structures of other dimensions can also be implemented by resistive memory arrays of corresponding dimensions. For example, a quasi-neural network structure of any number of layers can be realized by multiple resistive memory arrays. For example, in a quasi-neural network structure using a non-fully connected structure, the relationship between each node in the neuron layer circuit and the connected node of the previous neuron layer circuit can also be implemented by the resistive memory array. For example, in a quasi-neural network structure including an activation function circuit, the activation function circuit can be regarded as a circuit operated separately, so that the relationship between the two neuron layer circuits in the quasi-neural network structure can also be implemented by the resistive memory array.
[0062] The following further describes the implementation method of the physical unclonable function based on the neural network-like structure based on the use of a resistive memory array to implement the weight parameter matrix of the neural network-like structure.
[0063] In step S110, a voltage may be applied across the resistive memory in the resistive memory array to cause the formation and rupture of conductive filaments inside the resistive memory, thereby causing the conductivity of the resistive memory to change with the applied voltage.
[0064] For example, Figure 4 The figure shows a schematic diagram of the electrical characteristics of the resistive memory, wherein the horizontal axis is the voltage applied to the two ends of the resistive memory, and the vertical axis is the current value. Figure 4In the example shown, the resistive memory is a bipolar resistive memory with a 1T1R structure, and a DC sweep is performed on the resistive memory, specifically: the voltage at the source line end of the resistive memory is fixed at 0V, and then the voltage at the bit line end of the resistive memory is changed according to a cycle of 0V-2V-0V-(-2V)-0V. Figure 4 As shown, the conductance value of the resistive memory changes with the amplitude and polarity of the applied voltage. Specifically, when a forward bias is applied between the bit line terminal and the source line terminal of the resistive memory, the conductance value of the resistive memory increases; when a reverse bias is applied between the bit line terminal and the source line terminal of the resistive memory, the conductance value of the resistive memory decreases. For example, Figure 4 As shown, in the forward scanning stage, when the voltage is 1V, the current corresponding to the resistive memory is 100nA, but because the conductivity value of the resistive memory gradually increases, in the reverse scanning stage, when the voltage is 1V, the current corresponding to the resistive memory is 10uA.
[0065] It should be noted that in the embodiments of the present disclosure, the operation of reducing the conductance value of the operated resistive memory is called a reset operation, and the operation of increasing the conductance value of the operated resistive memory is called a set operation.
[0066] It should also be noted that Figure 4 The examples shown in the figure use a bipolar resistive memory and apply a voltage to both ends of the resistive memory by DC scanning, which are exemplary and not limiting to the present disclosure. For example, a voltage may be applied to both ends of the resistive memory by applying a pulse voltage. For example, a unipolar resistive memory may be used, and the conductance value of the unipolar resistive memory changes only with the amplitude of the applied voltage. For example, the voltage V applied to the word line end of the unipolar resistive memory is WL When it is high, the unipolar resistive memory is turned on. For the reset operation, the voltage V applied to the source line of the unipolar resistive memory is SL is the reset pulse, and the voltage V applied to the bit line of the unipolar resistive memory BL is a low level voltage (eg, ground); for a set operation, a voltage V applied to the bit line terminal of the unipolar resistive memory BL is the set pulse, the voltage V applied to the source line of the unipolar resistive memory SL is a low level voltage (eg, ground).
[0067] For example, Figure 5 The figure shows the current distribution diagram of multiple resistive memory devices, wherein the horizontal axis is the current value and the vertical axis is the number of resistive memory devices. Figure 5In the example shown, the same reset operation is performed on multiple unipolar resistive memories, and then a read operation is performed on the multiple unipolar resistive memories to read the current value. For example, the voltage for performing the reset operation is: V WL =5V, V BL =0V, V SL =3V; the voltage for performing the read operation is: V WL =4V, V BL =0.15V, V SL =0V. Figure 5 As shown, each resistive memory has a different conductance value after the same reset operation. This is because the formation and breaking of the conductive filaments inside the resistive memory are relatively random.
[0068] Therefore, in the embodiments of the present disclosure, the setting operation of the neural network-like structure can be implemented by performing a reset operation or a set operation on the resistive memory in the resistive memory array, thereby conveniently realizing the random setting of the weight parameters in the weight parameter matrix.
[0069] For example, in the implementation method provided by at least one embodiment of the present disclosure, a setting operation is performed on the quasi-neural network structure so that at least part of the weight parameters in the weight parameter matrix are randomly set, including: performing at least one reset operation on at least part of the resistive memory in the resistive memory array so that the conductance value of at least part of the resistive memory in the resistive memory array is randomly set. For example, the voltage for performing the reset operation is: V WL =4V, V BL =0V, V SL =2.2~2.8V. It should be noted that, generally, the higher the proportion (i.e., the number) of the resistive memory cells with randomly set conductance values in the resistive memory array, the better the randomness of the key value generated by the physical unclonable function. For example, a reset operation is performed on all the resistive memory cells in the resistive memory array so that the conductance values of all the resistive memory cells in the resistive memory array are randomly set.
[0070] For example, in the implementation method provided by at least one embodiment of the present disclosure, a setting operation is performed on the quasi-neural network structure so that at least part of the weight parameters in the weight parameter matrix are randomly set, including: performing at least one set operation on at least part of the resistive memory in the resistive memory array so that the conductance value of at least part of the resistive memory in the resistive memory array is randomly set. For example, the voltage for performing the set operation is: V WL =1.2~3V, V BL =2.5~3V, V SL=0V. It should be noted that, generally, the higher the proportion of the resistive memory with randomly set conductance values in the resistive memory array, the better the randomness of the key value generated by the physical unclonable function. For example, a set operation is performed on all the resistive memories in the resistive memory array so that the conductance values of all the resistive memories in the resistive memory array are randomly set.
[0071] For example, in the implementation method provided by at least one embodiment of the present disclosure, a setting operation is performed on the neural network structure so that at least part of the weight parameters in the weight parameter matrix are randomly set, including: performing a combination of at least one reset operation and at least one set operation on at least part of the resistive memory in the resistive memory array, so that the conductance value of at least part of the resistive memory in the resistive memory array is randomly set. For example, the voltage for performing the reset operation is: V WL =4V, V BL =0V, V SL =2.2~2.8V. For example, the voltage for performing a set operation is: V WL =1.2~3V, V BL =2.5~3V, V SL =0V. It should be noted that, generally, the higher the proportion of the resistive memory with randomly set conductance values in the resistive memory array, the better the randomness of the key value generated by the physical unclonable function. For example, a reset operation and a set operation are performed on all the resistive memories in the resistive memory array, so that the conductance values of all the resistive memories in the resistive memory array are randomly set.
[0072] For step S120, calculation may be performed based on the weight parameter matrix implemented by the resistive memory array and the input electrical signal to provide an output electrical signal.
[0073] For example, in the implementation method provided by at least one embodiment of the present disclosure, an input electrical signal is provided to a neural network-like structure, and the neural network-like structure provides an output electrical signal based on the input electrical signal and multiple weight parameters of a weight parameter matrix, including: applying a voltage to at least part of the resistive memory in the resistive memory array to output a current value, the voltage being the input electrical signal, and the current value being the output electrical signal.
[0074] For example, refer again to Figure 3A and Figure 3B For example, in some examples, Figure 3A The input electrical signal x1 of the neural network structure shown corresponds to the value 0, and x2 corresponds to the value 1, then Figure 3B The voltage V1 on the word line end of the resistive memory array shown in the figure can be 0V, and the voltage V2 on the word line end can be a read voltage (i.e., only Figure 3BThe resistive memory of the second row of the resistive memory array shown in the figure is applied with a voltage), for example, the read potential may be 0.15V. In this example, based on V2, g 21 and g 22 The currents I1 and I2 at the bit line terminals are calculated. For example, in some other examples, Figure 3A The input electrical signal x1 of the neural network structure shown corresponds to the value 1, and x2 corresponds to the value 1, then Figure 3B The voltage V1 on the word line terminal of the resistive memory array shown in the figure can be a read voltage, and the voltage V2 on the word line terminal can be a read voltage (ie, Figure 3B The resistive memory array shown in the figure has a voltage applied to all resistive memory cells), for example, the read potential may be 0.15 V. In this example, the voltage may be applied to all resistive memory cells of the resistive memory array shown in the figure, for example, the read potential may be 0.15 V. 11 , g 12 , g 21 and g 22 The currents I1 and I2 at the bit line terminals are calculated.
[0075] It should be noted that the number of current values is related to the number of key values corresponding to each stimulus response pair. For example, in one example, each stimulus response pair is expected to correspond to a 128-bit key value, and assuming that each bit of the key value will be determined based on a current value, a neural network-like structure can be arranged so that 128 current values can be output after applying a voltage to at least part of the resistive memory in the resistive memory array.
[0076] For step S130, the output current value may be processed to obtain a key value of a physical unclonable function.
[0077] For example, in an implementation method provided by at least one embodiment of the present disclosure, determining the key value of a physical unclonable function based on an output electrical signal includes: comparing the output current value with a reference current value, and outputting the key value of the physical unclonable function based on the comparison result.
[0078] For example, a comparator may be used to compare the output current value with a reference current value. For example, if the output current value is greater than the reference current value, the key value of the output physical unclonable function is 1; if the output current value is less than or equal to the reference current value, the key value of the output physical unclonable function is 0. Alternatively, if the output current value is greater than the reference current value, the key value of the output physical unclonable function is 0; if the output current value is less than or equal to the reference current value, the key value of the output physical unclonable function is 1. For example, the reference current value may be obtained based on experience, depending on the actual situation, and the embodiments of the present disclosure are not limited to this.
[0079] It should be noted that other technical means may also be used to process the output current value to obtain the key value of the physical unclonable function, and the embodiments of the present disclosure are not limited to this.
[0080] For example, Figure 6 The distribution diagram of the key value generated by the physical unclonable function according to the embodiment of the present disclosure is illustrated. Figure 6 In the figure, the key values of the physical unclonable function corresponding to 80 stimulus-response pairs are illustrated. Each stimulus-response pair outputs a 128-bit key value, where a black dot indicates a key value of 0 and a white dot indicates a key value of 1. Figure 6 As shown, the black points and the white points are evenly distributed, that is, the generated key value has good randomness.
[0081] Therefore, the physical unclonable function based on a neural network-like structure proposed in the embodiment of the present disclosure fully utilizes the randomness of the weight parameters in the weight parameter matrix, so that the generated key value satisfies the random distribution.
[0082] It should be noted that, in the embodiments of the present disclosure, the process of the implementation method of the physical unclonable function based on the neural network-like structure provided by the above-mentioned various embodiments of the present disclosure may include more or fewer operations, and these operations may be executed sequentially or in parallel. Although the process of the implementation method of the physical unclonable function based on the neural network-like structure described above includes multiple operations that appear in a specific order, it should be clearly understood that the order of multiple operations is not restricted. The implementation method of the physical unclonable function based on the neural network-like structure described above can be executed once, or it can be executed multiple times according to predetermined conditions.
[0083] The implementation method of the above-mentioned physical unclonable function provided by the embodiment of the present disclosure achieves a huge number of stimulus-response pairs with a relatively small hardware overhead by randomly setting the weight parameters in the neural network-like structure, and utilizes the neural network-like structure to improve the data complexity of the stimulus-response pairs by utilizing the complex connection relationship within the neural network-like structure, thereby improving the anti-modeling attack capability of the implemented PUF.
[0084] Figure 7 A schematic block diagram of a device for implementing a physical unclonable function based on a neural network-like structure provided by at least one embodiment of the present disclosure. Figure 7 As shown, the device 200 for implementing the physical unclonable function includes: a setting circuit 210, an input circuit 220, a storage and calculation circuit 230 and an output circuit 240.
[0085] The storage and calculation circuit 230 includes a neural network-like structure, which includes multiple neuron layer circuits and a weight parameter matrix between the multiple neuron layer circuits. For example, the implementation method of the neural network-like structure can refer to the relevant description of the neural network-like structure in the implementation method, which will not be repeated here.
[0086] The setting circuit 210 is configured to perform a setting operation on the storage calculation circuit 230 so that at least part of the weight parameters in the weight parameter matrix are randomly set. For example, the setting circuit 210 can implement step S110, and its specific implementation method can refer to the relevant description of step S110, which will not be repeated here.
[0087] The input circuit 220 is configured to provide an input electrical signal to the storage computing circuit 230. For example, the input circuit 220 can implement the process of providing the input electrical signal in step S120, and its specific implementation method can refer to the relevant description in the implementation method, which will not be repeated here.
[0088] The storage calculation circuit 230 is configured to provide an output electrical signal based on the input electrical signal and multiple weight parameters of the weight parameter matrix. For example, the storage calculation circuit 230 can implement the process of providing an output electrical signal based on the input electrical signal and multiple weight parameters of the weight parameter matrix in step S120. The specific implementation method can refer to the relevant description in the implementation method, which will not be repeated here.
[0089] The output circuit 240 is configured to determine the key value of the physical unclonable function based on the output electrical signal. For example, the output circuit 240 can implement step S130, and its specific implementation method can refer to the relevant description of step S130, which will not be repeated here.
[0090] For example, in the implementation device provided by at least one embodiment of the present disclosure, the storage calculation circuit 230 includes a resistive memory array, the resistive memory array serves as a weight parameter matrix of a neural network-like structure, the conductance values of multiple resistive memories in the resistive memory array correspond to multiple weight parameters of the weight parameter matrix, and each resistive memory in the resistive memory array includes a first end connected to the previous neuron layer circuit and a second end connected to the next neuron layer circuit. For example, the implementation method of the resistive memory array as a weight parameter matrix of a neural network-like structure can refer to the relevant description in the implementation method, which will not be repeated here.
[0091] For example, in the implementation device provided by at least one embodiment of the present disclosure, the first end of the first part of the resistive memory array is connected to the input circuit to receive the input electrical signal, and the second end of the second part of the resistive memory array is connected to the output circuit to output the output electrical signal. For example, the signal connection of the resistive memory array can refer to Figure 3B The relevant description of the example is not repeated here. It should be noted that in the embodiment of the present disclosure, the first part and the second part are only for the convenience of description, and the resistive memory array may also include other parts of the resistive memory, and the resistive memory included in the first part of the resistive memory and the second part of the resistive memory may be the same or different, depending on the specific actual situation. For example, Figure 3B In the example shown, the first portion of resistive memories may include all four resistive memories, and the second portion of resistive memories may also include all four resistive memories.
[0092] For example, in the implementation device provided by at least one embodiment of the present disclosure, the setting circuit 210 is configured to: perform at least one reset operation on at least part of the resistive memory in the resistive memory array, so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set. For example, the setting operation implemented by the setting circuit 210 can refer to the relevant description in the implementation method, and will not be repeated here.
[0093] For example, in the implementation device provided by at least one embodiment of the present disclosure, the setting circuit 210 is configured to: perform at least one set operation on at least part of the resistive memory in the resistive memory array, so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set. For example, the setting operation implemented by the setting circuit 210 can refer to the relevant description in the implementation method, and will not be repeated here.
[0094] For example, in the implementation device provided by at least one embodiment of the present disclosure, the setting circuit 210 is configured to: perform a combination of at least one reset operation and at least one set operation on at least part of the resistive memory in the resistive memory array, so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set. For example, the setting operation implemented by the setting circuit 210 can refer to the relevant description in the implementation method, and will not be repeated here.
[0095] For example, in the implementation device provided by at least one embodiment of the present disclosure, the input circuit 220 is configured to apply a voltage to at least part of the resistive memory in the resistive memory array to output a current, and the storage calculation circuit 230 is configured to obtain and output the current value of the current, the voltage is the input electrical signal, and the current value is the output electrical signal. For example, the operations implemented by the input circuit 220 and the storage calculation circuit 230 can refer to the relevant description in the implementation method, and will not be repeated here.
[0096] For example, in the implementation device provided by at least one embodiment of the present disclosure, the output circuit 240 is configured to: compare the output current value with the reference current value, and output the key value of the physical unclonable function based on the comparison result. For example, the way in which the output circuit 240 generates the key value can refer to the relevant description in the implementation method, and will not be repeated here.
[0097] For example, in some other examples, the device 200 for implementing the physical unclonable function may also include one or more processors (not shown in the figure) and a memory (not shown in the figure).
[0098] For example, the processor may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, may be a general-purpose processor or a dedicated processor, and may control other components in the implementation device 200 to perform desired functions. The memory may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on a computer-readable storage medium, and the processor may run the program instructions to implement the functions (implemented by the processor) in the embodiment of the present disclosure and / or other desired functions, such as controlling the execution of setting operations, etc. Various applications and various data, such as reference current values and various data used and / or generated by applications, etc., may also be stored in the computer-readable storage medium.
[0099] It should be noted that, for the sake of clarity and brevity, the embodiment of the present disclosure does not provide all the components of the device for implementing the physical unclonable function 200. To implement the necessary functions of the device for implementing the physical unclonable function 200, those skilled in the art may provide and set other components not shown according to specific needs, and the embodiment of the present disclosure does not limit this.
[0100] Regarding the technical effects of the device 200 for implementing a physical unclonable function in different embodiments, reference may be made to the technical effects of the method for implementing a physical unclonable function provided in the embodiments of the present disclosure, which will not be described in detail here.
[0101] There are a few points to note:
[0102] (1) The drawings of the embodiments of the present disclosure only relate to the structures related to the embodiments of the present disclosure, and other structures may refer to the general design.
[0103] (2) In the absence of conflict, the embodiments of the present disclosure and the features therein may be combined with each other to obtain new embodiments.
[0104] The above description is merely an exemplary embodiment of the present disclosure and is not intended to limit the protection scope of the present disclosure. The protection scope of the present disclosure is determined by the appended claims.
Claims
1. A method for realizing a physical unclonable function based on a neural network-like structure, wherein the neural network-like structure includes a plurality of neuron layer circuits and a weight parameter matrix between the plurality of neuron layer circuits, the method comprising: Performing a setting operation on the neural network-like structure so that at least part of the weight parameters in the weight parameter matrix are randomly set; Providing an input electrical signal to the neural network-like structure, wherein the neural network-like structure provides an output electrical signal based on the input electrical signal and a plurality of weight parameters of the weight parameter matrix; determining a key value of the physical unclonable function based on the output electrical signal, The method further comprises: The weight parameter matrix of the neural network structure is realized by using a resistive memory array. The conductance values of the multiple resistive memory cells in the resistive memory array correspond to the multiple weight parameters of the weight parameter matrix, respectively. The neural network-like structure is set so that at least part of the weight parameters in the weight parameter matrix are randomly set, including: Performing at least one reset operation on at least part of the resistive memory in the resistive memory array, so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set; or Performing at least one set operation on at least part of the resistive memory in the resistive memory array, so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set; or A combination of at least one reset operation and at least one set operation is performed on at least part of the resistive memory in the resistive memory array, so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set.
2. The method according to claim 1, wherein: Each resistive memory in the resistive memory array includes a first end connected to a previous neuron layer circuit and a second end connected to a next neuron layer circuit.
3. The method according to claim 2, wherein: A fully connected structure is adopted between the multiple neuron layer circuits.
4. The method according to claim 2, wherein: An activation function circuit is also included between the multiple neuron layer circuits.
5. The method according to any one of claims 2 to 4, wherein: Providing an input electrical signal to the neural network-like structure, wherein the neural network-like structure provides an output electrical signal based on the input electrical signal and a plurality of weight parameters of the weight parameter matrix, including: applying a voltage to at least part of the resistive memory in the resistive memory array to output a current value, Among them, the voltage is the input electrical signal, and the current value is the output electrical signal.
6. The method according to claim 5, wherein: Determining a key value of the physical unclonable function based on the output electrical signal includes: The output current value is compared with a reference current value, and a key value of the physical unclonable function is output based on the comparison result.
7. The method according to claim 1, wherein: The reset operation reduces the conductance value of the operated resistive memory, and the set operation increases the conductance value of the operated resistive memory.
8. A device for implementing a physical unclonable function based on a neural network-like structure, the device for implementing a physical unclonable function comprising: Setting circuit, input circuit, storage calculation circuit and output circuit, The storage and calculation circuit includes the neural network-like structure, and the neural network-like structure includes a plurality of neuron layer circuits and a weight parameter matrix between the plurality of neuron layer circuits. The setting circuit is configured to perform a setting operation on the storage calculation circuit so that at least part of the weight parameters in the weight parameter matrix are randomly set; The input circuit is configured to provide an input electrical signal to the storage computing circuit; The storage and calculation circuit is configured to provide an output electrical signal based on the input electrical signal and a plurality of weight parameters of the weight parameter matrix; The output circuit is configured to determine a key value of the physical unclonable function based on the output electrical signal, The storage computing circuit includes a resistive memory array, The resistive memory array serves as a weight parameter matrix of the neural network-like structure. The conductance values of the plurality of resistive memory cells in the resistive memory array correspond to the plurality of weight parameters of the weight parameter matrix, respectively. The setting circuit is configured to: perform at least one reset operation on at least part of the resistive memory in the resistive memory array, so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set; or Performing at least one set operation on at least part of the resistive memory in the resistive memory array, so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set; or A combination of at least one reset operation and at least one set operation is performed on at least part of the resistive memory in the resistive memory array, so that the conductance values of at least part of the resistive memory in the resistive memory array are randomly set.
9. The implementation device according to claim 8, wherein: Each resistive memory in the resistive memory array includes a first end connected to a previous neuron layer circuit and a second end connected to a next neuron layer circuit.
10. The implementation device according to claim 9, wherein: The first end of the first part of the resistive memory array is connected to the input circuit to receive the input electrical signal. The second end of the second part of the resistive memory of the resistive memory array is connected to the output circuit to output the output electrical signal.
11. The implementation device according to claim 9, wherein: The input circuit is configured to apply a voltage to at least part of the resistive memory in the resistive memory array to output a current, and the storage calculation circuit is configured to obtain and output a current value of the current. Among them, the voltage is the input electrical signal, and the current value is the output electrical signal.
12. The implementation device according to claim 11, wherein: The output circuit configuration is: The output current value is compared with a reference current value, and a key value of the physical unclonable function is output based on the comparison result.
