A Method for Generating Truly Random Numbers

Abstract

The invention relates to a true random number generator (TRNG) method based on Double Data Rate (DDR) that is intended for use in generating keys and one-time pads (OTPs) in various applications in cryptography and secure communication. It features low resource and power consumption, high entropy in cryptographic and security components, and resistance to machine learning attacks.

Description

1 TARIFF A Method for Generating Truly Random Numbers TECHNICAL AREA 5 The invention provides keys and disposable tape for various applications in cryptography and secure communications. Intended for use in One Time Pad (OTP) production, DDR (Double Data Rate) (Ratio) based low resource and power consumption, in cryptographic and security units. a 10 that has high entropy and is resistant to machine learning attacks This relates to the True Random Number Generator (TRNG) method. PREVIOUS TECHNIQUE The True Random Number Generator (TRNG) attributes physical events to 15 It is a system that generates random numbers based on a traditional pseudo-random number generator. Unlike Pseudo-Random Number Generator (PRNG) algorithms, TRNGs generate numbers. It relies on non-deterministic, or unpredictable, processes to produce it. This Processes are generally random events in the physical world, such as electrical noise, atmospheric noise. Events such as radio waves or atomic decay can occur. The main advantage of TRNGs is that 20 The numbers they generate are completely random, which makes them suitable for security applications. This makes it ideal for encryption systems, simulations, and more. However, TRNGs generally operate slower than PRNGs and require additional hardware in systems. The need for certain requirements may arise. Therefore, depending on the areas of use... It has advantages and disadvantages. 25 Modern TRNGs rely on physical processes or natural processes instead of computer-based algorithms. These are random numbers generated by events. These numbers are based on a specific pattern or formula. Because they are not generated in this way, they are truly random. Cryptography and security fields. This is one of the most important application areas of TRNGs. Secure key generation, session 30 keys, initialization vectors for encryption algorithms, and other cryptographic protocols. Reliable random numbers are required. True randomness enhances cryptographic security and predictability. It eliminates the possibility of feasibility. Traditional TRNGs are problematic due to the complexity of the application, high energy consumption, and environmental impacts. It encounters problems such as sensitivity to factors. New developments address various entropy issues. Although they are geared towards exploring resources, they generally have high entropy and low hardware specifications. 2 Striking a balance between requirements and features like resistance to attacks is difficult. The effectiveness of TRNGs is critical, especially where ensuring security and unpredictability is essential. It is of great importance in these contexts. Traditional TRNGs use thermal noise to generate randomness, It utilizes physical phenomena such as jitter or quantum effects in electronic circuits. However, these methods often have limitations in terms of implementation complexity, power consumption, and environmental changes. It has challenges related to sensitivity. Recent developments in TRNG design come from various sources. It has explored innovative approaches to using disorder. These TRNG schemes, various Digital Clock Manager (DCM) based on ring oscillators It extends to those who use it. Despite these developments, current TRNG schemes are generally High entropy, low hardware overhead, and machine learning (ML) 10 It falls short in addressing all the desired qualities, such as resistance to attacks. As a result of research conducted in the literature, the application numbered “2023 / 016218” and “REAL” A Turkish patent application titled "RANDOM NUMBER GENERATOR" has been found. The subject of the application is a digital chaotic oscillating autonomous Boolean network as a source of entropy 15 It relates to devices that generate truly random numbers. However, The aforementioned application discusses key and single applications in cryptography and secure communication. Used for the production of disposable ribbons, it is DDR-based and has low resource and power consumption. cryptographic and security units have high levels of disorder and machine learning. 20 is an indication of a truly random number generator method that is resistant to attacks. None have been encountered. Ultimately, the problems mentioned above, which cannot be solved with current technology, are the subject of this technical analysis. This has made it necessary to make an innovation in the field. BRIEF DESCRIPTION OF THE INVENTION 25 The present invention aims to eliminate the aforementioned disadvantages and introduce new technologies to the relevant technical field. It is related to a true random number generator method designed to bring advantages. The main purpose of the invention is to provide key and single 30 applications in cryptography and secure communication. Used for the production of disposable ribbons, DDR-based, low resource and power consumption, and cryptographic and security units with high disorganization and machine learning The goal is to create a truly random number generator that is resistant to attacks. All the objectives mentioned above and those that will emerge from the detailed explanation below are 35 The present invention aims to realize the entropy source for a random number generator. 3 At least one entropy cell used, De-, detects randomness in the bit sequence given as input. The biasing unit collects the bits extracted by the de-biasing unit and displays them in bytes. In various applications in cryptography and secure communication, involving a sample unit that produces output. Low-source and low-power DDR-based systems used for the production of switches and disposable ribbons. machine 5, which has high disorganization in consumption, cryptographic and security units. It is a true random number generator method that is resistant to learning attacks, Its features include an entropy cell, a de-biasing unit, and a sampling unit integrated into the system. Starting the generation of random numbers along with its operation, the aforementioned entropy cell The entropy sources within it begin to generate data entropy-input data, entropy If the cell is a ring oscillator, its structure inherently generates 1 bit on its own, 10 If the entropy source is DDR, then external inputs are provided from RAM and The 3-bit value generated is used to start the XOR operation, and the 1-bit output of the XOR operation is... When the value is given as input to the De-Biasing unit, the entropy of the De-Biasing unit is determined. The system monitors the randomness of its state based on XORed data coming from the cell, De- In the biasing unit, randomness is increased using a randomness extractor, and the output bit is 15. It is stored until 1 byte of data is collected by giving it to the sample unit, and 1 byte This involves providing a new random number as output when the event occurs. The best way to utilize the advantages of the existing invention, together with its structure and additional elements. For it to be understood, it must be evaluated together with the figures explained below. 20 BRIEF DESCRIPTION OF THE FIGURES Figure 1 is a representative illustration of the original neoTRNG design. Figure 2 shows a representation of a random number generator with 3 ring oscillators and different numbers of inverters. 25 It is a representation. Figure 3 shows a random number with different numbers of inverters, having 2 ring oscillators and 1 DDR input. It is a representative illustration of the generator. Figure 4 shows a random number with different numbers of inverters, 1 ring oscillator and 2 DDR inputs. It is a representative illustration of the generator. 30 Figure 5 is a representative illustration of a random number generator with 3 DDR inputs. Figure 6 is a representative illustration of a true random number generator method that is the subject of the invention. The drawings do not necessarily need to be scaled and are necessary for understanding the invention. Details that are not present may have been overlooked. Furthermore, at least to a large extent, 35 Elements that are identical or at least have substantially identical functions are numbered the same. It is shown. 4 REFERENCE NUMBERS 1. Entropy cell 2. De-biasing unit 5 3. Sample unit 1001. The system in which the entropy cell, de-biasing unit, and sampling unit are integrated. Starting the random number generation process when the system is activated 1002. Entropy sources within the entropy cell: data entropy-input data 10 to start producing 1003. If the entropy cell is a ring oscillator, its structure inherently... if it generates 1 bit for itself, and the entropy source is DDR, then it receives it from outside. The inputs are provided from RAM and the XOR operation is performed using the generated 3-bit value. start 15 The 1-bit value, which is the output of the 1004th XOR operation, is input into the De-Biasing unit. to be given as According to the XORed data from the entropy cell of the 1005th De-Biasing unit: the system's randomness status 1006. In the De-Biasing unit, 20 appears to be source-independent and evenly distributed. entropy cell output of the randomness extractor that produces a highly random output The process involves applying and providing the output bit to the sample unit, resulting in 1 byte of data. It is stored until collected, and when it reaches 1 byte, a new random number is output. giving DETAILED DESCRIPTION OF THE INVENTION This detailed explanation describes a true random number generator method that is the subject of the invention. with examples that will not have any limiting effect on a better understanding of the subject It is explained. 30 A true random number generator method, entropy cell (1), De-Biasing unit (2), sample It includes unit (3). Entropy cell (1) as the entropy source for the random number generator. It is used. In our invention, we based our design on the DDR RAM entropy source, although it is ring-based. In other versions of our invention in the oscillator, 35 is used as an auxiliary entropy source. It is used. It is a hardware element used within architecture. De- The biasing unit (2) detects randomness in the bit sequence given as input. Method in the invention preferably as a John Von Neumann randomness extractor (John Von Neumann Randomness The Extractor has been selected. It is included as a hardware element used within the architecture. The sampling unit (3) determines the randomness distribution of the bits provided by the De-Biasing unit (2). It is a unit that increases and provides byte-width output. It is a hardware element used within the architecture. It is listed as follows. 5 In a true random number generator method, the neoTRNG architecture given in [1] (See Figure 1) lightweight DDR RAM resistant to machine learning attacks and ring oscillator (RO) based TRNG design is implemented by presenting the existing random number Solutions are offered to the problems that generators have. These problems are solved in 10 steps. Using DDR RAM instead of ring oscillator as entropy source on neoTRNG Hardware update has been provided. The invention design is FPGA (Field Programmable Gate Array). - Implemented on Programmable Gate Arrays in the field and in terms of resource consumption. To observe that it is a lightweight, FPGA source was used via the Vivado application. Consumption was achieved by obtaining the result of synthesis. 15 of the data obtained from all the schemes presented in our invention. The results are compared with other studies in the academic literature in Table 1. In addition, the input bits given to the entropy cell (1) via the FPGA and the corresponding output By collecting random bits, the resistance of the machine to machine learning attacks can be observed. A dataset has been created. The system should generate an 8-bit output from the created dataset. Therefore, each sample in the dataset corresponds to an 8-bit output for every 24-bit entropy source. They are stored as follows: For each design given in Figure 1, Figure 2, Figure 3, Figure 4 and Figure 5. Separate datasets were created. In the design given in Figure 2, different numbers of inverters were used. Three RO units are used as entropy sources. With this design, the entropy ratios are 3-5-7, 3-5-9, and 3-7-9 respectively. Three different RO-based TRNG design schemes with varying numbers of inverters have been created. (Figure 3 shows this). In the given design, 2 RO and 1 DDR RAM data are used as entropy sources, and this 25 The design allows for 5 different RO and DDR RAM configurations with inverter counts of 3-5, 3-7, 3-9, 5-7, and 5-9 respectively. A hybrid TRNG design scheme based on 2 DDR RAM data and 1 RO was used as the entropy source. With this design, the number of inverters was 3, 5, 7, and 9 respectively. Four different hybrid TRNG design schemes based on RO and DDR RAM have been created. (Figure) In the design given in section 5, 3 DDR RAM data is used as the entropy source, resulting in 30 DDR RAM. A TRNG design scheme based on the above has been created. When all designs are considered, there are a total of 13. For different TRNG design schemes; each with 1 million bits of generated random numbers and corresponding values. Thirteen different datasets were created to include entropy data. Each random number Machine learning algorithms applied to datasets generated by the generator Polynomial Regression, Linear Regression, Decision Tree, Random Forest, 35 Radial SVM, Polynomial SVM, Linear SVM, K-Means, Neural Networks, and Long-Term and Short-Term Memory (LSTM - Long Short Term Memory). Table 2 shows the applied machine learning. 6 It includes the hyperparameters of the algorithms, and Table 3 shows the machine learning algorithms. The results of the attacks are shown. For hyperparameters not listed in Table 2, the default is... The values ​​have been used. When resource consumption on the FPGA is examined, the 5 proposed in this invention in terms of LUT and FF is considered. The designs have been proven to consume fewer resources compared to other studies in the literature (See). Table 1). Furthermore, the invention offers the highest speed in terms of operating frequency on the FPGA. It has been observed that it has the capacity to operate. However, in terms of power consumption... It appears to consume more power than some of the studies we mentioned. The TRNG design method given in a true random number generator method is used for the machine. learning attacks (a prerequisite for TRNG design to be accepted) (one of the conditions) reliability has also been tested. As a result of machine learning training. The accuracy and F1-score metrics for all algorithms are close to 0. Because it contains multiple values, attack algorithms were unable to find a relationship between the generated values. 15 And they haven't been able to learn it. So the TRNG method proposed within the scope of the invention is currently one of the most important. It is not vulnerable to machine learning attacks, which are one type of attack. A true random number generator method is readily available in modern computing systems. By utilizing the entropy provided by DDR RAM, a resource that offers low resource consumption but still at 20%, It eliminates the need for a secure TRNG design. The invention proposes the following: This method / approach not only guarantees high entropy and robustness, but also by minimizing additional hardware requirements over time, it makes it one of today's TRNGs. It makes it a compelling solution for their needs. The original neoTRNG design [1] is based on a true random number generator method. Four different new TRNG design methods were developed (See Figure 2, Figure 3, Figure 4, Figure 5). It has been suggested. Steps in a true random number generator method: 30  Randomly occurs when the system in which the invention is integrated is started up (booted up). Number generation begins (1001).  The 3 entropy sources located inside the entropy cell (1) start generating data. (1002). If the entropy source is RO, it will generate 1 bit on its own by its nature, If the entropy source is DDR, the input bit will be provided externally via RAM. 35 The generated 3-bit value is then subjected to an XOR operation (1003). 7  The 1-bit value, which is the output of the XOR operation, is used as input to the De-Biasing unit (2). (1004). De-Biasing unit (2) is given to XORed data from entropy cell (1) According to this, it monitors the randomness of the system (1005).  In the De-Biasing unit (2), bits that have received confirmation for randomness are sent to the sampling unit (3) data of 1 byte length is collected (the scheme basically involves collecting 3 bits of data in each round). It produces a 1-bit output in response to an entropy input, with a target of 1 byte (8 bits) in the design. Since the output is of a certain length, running the scheme 8 turns, in other words, 3 bits of entropy. (Input * 8 turns = 24 bits of entropy input required) stored and output bit length When the total is 1 byte, a new random number is given as output (1006). The expression XORing refers to the XORing of bits generated by entropy cells (1). This means combining entropy cells (1), real random number generator (TRNG) These are components that generate numbers based on physical randomness in systems like these. Each of these cells... One generates a bit (0 or 1) based on physical random events. These bits often combine. The resulting random number is processed to obtain a longer, more complex random number. XOR 15 The exclusive OR (indirect) operation compares each bit, and the result is 1 if the two bits are different, and 1 if they are the same. It is a logic operation that returns a result of 0. In this operation, for example, the two bits could be: Bit1 = 1, Bit2 = 0 → XOR result = 1 Bit1 = 0, Bit2 = 1 → XOR result = 1 20 Bit1 = 1, Bit2 = 1 → XOR result = 0 Bit1 = 0, Bit2 = 0 → XOR result = 0 In this case, the different bits from the entropy cells (1) are XORed together to form a single bit. A sequence of bits is created. This bit sequence is used to make the randomness more homogeneous and to reduce low-quality 25 bits. It is processed in order to eliminate random bits. This process is De-Biasing (2) (bias (de-encryption) unit then processes it and makes the randomness more reliable, like encryption. This ensures that it becomes usable in applications. In short, the term XORing refers to various entropy Combining the random bits from the cells (1) and combining them by XOR operation This means that this process is generally used to improve the quality of random bits. It is used. A true random number generator method is the neoTRNG (Figure 1) true random number generator. It is a simply structured TRNG based on a ring oscillator and DDR RAM, using its architecture as a basis. A true random number generator method consists of three hardware elements; entropy 35 Entropy cell (1), De-Biasing unit (2) and sampling unit (3). Entropy cell (1) module, TRNG 8 It serves as a source of entropy for each entropy cell (1) to produce entropy. It uses a ring oscillator and / or DDR RAM. Three entropy cells (1) each producing one bit of output. In the original design of neoTRNG (Figure 1), the ring oscillators in these cells were 5, respectively. 7 and 9 contain reversing elements. In this invention, ring oscillators are reversed in different combinations. Combine different numbers of DDR RAM outputs as an entropy source (Figure 2, Figure 3, Figure 4 and Figure 5) 5 It is also used as. The bits produced by these entropy cells (1) are XORed together. and the result is fed to the De-Biasing unit (2). The De-Biasing unit (2) usually contains simply A randomness extractor, referred to as an "extractor," is involved. Randomness extractors a smooth, seemingly source-independent expression applied to the output from a weak entropy source. It is a structure that produces a highly random output that appears scattered. Our invention contains 10 John von Neumann randomness extractor, remaining faithful to the original work. The John von Neumann randomness extractor was used. The entropy extractor of cells (1) Monitoring whether the streamed bits on the XORed outputs disrupt the randomness of the process. It does this and produces an output when it satisfies the state of randomness. The system does not disrupt its randomness. The detected bits are used by the sampling unit (3) to generate byte-sized random numbers 15 It is sampled and serialized. The sample unit (3) also shows the spectral distribution of randomness. It includes a simple post-processing structure for development. Four different TRNG designs proposed within the scope of a true random number generator method. Comparisons of resource consumption related to the method and 20 obtained against machine learning attacks. The experimental results obtained are detailed in the tables below. Experimental Tables: Study Year Device LUT FF Frequency (MHz) Strength Consumption (mW) Entropy Source Piscopo et al. [2] 2024 Artix-7 106 194 324.7 99 Ring oscillator Spartan-7 98 194 230.3 104 Ring oscillator Oliveira et al. [3] 2023 Zynq-7000 73 6 50 N / A Ring oscillator Frustaci et al. [4] 2023 Zynq-7000 38 121 300 119 DCM Wang et al. [5] 2024 Spartan-6 42 21 N / A 112 NXRO 9 Artix-7 NXRO Grujić et al. [6] 2022 Zynq-7000 32 55 9.514 Ring oscillator Original neoTRNG 2023 Zynq-7000 45 46 333.34 107 Ring oscillator 3DDR TRNG Available Invention Zynq-7000 14 40 333.34 109 DRAMA Based on 2DDR- 1RO TRNG 3 Reverse eater Available Invention Zynq-7000 25 24 333.34 114 DRAMA Based on 2DDR- 1RO TRNG Reverse eater Available Invention Zynq-7000 27 26 333.34 113 DRAMA Based on 2DDR- 1RO TRNG 7 Reverse eater Available Invention Zynq-7000 29 28 333.34 114 DRAMA Based on 2DDR- 1RO TRNG 9 Reverse eater Available Invention Zynq-7000 31 30 333.34 114 DRAMA Based on 1 DDR- 2RO TRNG 3-5 Reverse eater Available Invention Zynq-7000 31 31 333.34 113 DRAMA Based on 1 DDR- 2RO TRNG 3-7 Reverse eater Available Invention Zynq-7000 33 33 333.34 114 DRAMA Based on 1 DDR- 2RO TRNG 3-9 Reverse eater Available Invention Zynq-7000 33 33 333.34 114 DRAMA Based on 1 DDR- 2RO TRNG 5-7 Reverse eater Available Invention Zynq-7000 35 35 333.34 113 DRAMA Based on 1 DDR- 2RO TRNG 5-9 Reverse eater Available Invention Zynq-7000 37 37 333.34 114 DRAMA Based on 3 RO TRNG 3-5-7 Reverse eater Available Invention Zynq-7000 43 44 333.34 107 Ring oscillator 3 RO TRNG 3-5-9 Reverse eater Available Invention Zynq-7000 41 42 333.34 107 Ring oscillator 3 RO TRNG 3-7-9 Reverse eater Available Invention Zynq-7000 39 40 333.34 107 Ring oscillator Table 1. All Random Number Generators Found in the Literature and Included in the Invention. Comparison of Resource Consumption Algorithm Hyperparameter Polynomial Regression "degree": 3 Decision Tree "criterion":"entropy", "random_state": 0 Random Forest "n_estimators": 2, "random_state":0, "fault_tolerance": 0.5 Radial SVM "C": 10, "gamma": 0.001, "kernel":"rbf" K-means "n_clusters":10, random_state":0 Neural Network "input_dim": 24, "hidden_layers": {[}64, 64{]}, "learning_rate": 0.001, "activation":"relu", "output_activation":"sigmoid", "loss_function":"binary_crossentropy" 11 LSTM "input_shape": (24, 1), "hidden_layers": {[}64, 64, 32{]}, "learning_rate": 0.001, "activation":"sigmoid", "output_activation":"sigmoid", "loss_function":"binary_crossentropy" Table 2. Algorithms and Hyper-Analysis Applied for Machine Learning Attacks. Parameters Algorithm Metric Design 3 DD R TR NG 2DDR-1RO TRNG 1DDR-2RO TRNG 3RO TRNG 3 5 7 9 3-5 3-7 3-9 5-7 5-9 3- 5-7 3- 5-9 3- 7-9 5-7-9 ( neoT RNG ) Polynomial Regression True UK 0.0 033 0.0 04 0.0 033 0.0 042 0.0 037 0.0 029 0.0 036 0.00 35 0.0 04 0.0 04 0.0 036 0.0 033 0.0 035 0.002 8 AUC 0.4 64 0.4 698 0.4 962 0.4 891 0.4 88 0.5 1 0.4 863 0.49 04 0.5 155 0.4 687 0.5 194 0.4 983 0.4 929 0.480 2 Sensitive ness 0.0 034 0.0 04 0.0 033 0.0 042 0.0 037 0.0 029 0.0 036 0.00 35 0.0 04 0.0 04 0.0 036 0.0 033 0.0 035 0.002 8 Definitely k 0.0 034 0.0 04 0.0 033 0.0 042 0.0 037 0.0 029 0.0 036 0.00 35 0.0 04 0.0 04 0.0 036 0.0 033 0.0 035 0.002 8 F1 0.0 034 0.0 04 0.0 033 0.0 042 0.0 037 0.0 029 0.0 036 0.00 35 0.0 04 0.0 04 0.0 036 0.0 033 0.0 035 0.002 8 Kappa - 0.0 005 ~0 - 0.0 005 0.0 003 4 - 0.0 001 - 0.0 009 - 0.0 002 - 0.00 03 ~0 0.0 001 - 0.0 002 - 0.0 005 - 0.0 004 - 0.001 12 MCC - 0.0 005 ~0 - 0.0 005 0.0 003 4 - 0.0 001 - 0.0 009 - 0.0 002 - 0.00 03 ~0 0.0 001 - 0.0 002 - 0.0 005 - 0.0 004 - 0.001 Logistics Regression True UK 0.0 039 0.0 041 0.0 033 0.0 039 0.0 041 0.0 035 0.0 04 0.00 45 0.0 038 0.0 041 0.0 039 0.0 033 0.0 042 0.003 6 AUC 0.5 138 0.5 067 0.4 822 0.5 065 0.4 826 0.4 979 0.4 735 0.50 69 0.5 016 0.5 425 0.5 008 0.4 839 0.5 181 0.507 7 Sensitive ness 0.0 039 0.0 041 0.0 033 0.0 039 0.0 041 0.0 035 0.0 04 0.00 45 0.0 038 0.0 041 0.0 039 0.0 033 0.0 042 0.003 6 Definitely k 0.0 014 0.0 041 0.0 033 0.0 039 0.0 041 0.0 035 0.0 04 0.00 45 0.0 038 0.0 041 0.0 039 0.0 033 0.0 042 0.003 6 F1 0.0 014 0.0 041 0.0 033 0.0 039 0.0 041 0.0 035 0.0 004 0.00 45 0.0 038 0.0 041 0.0 039 0.0 033 0.0 042 0.003 6 Kappa - 0.0 024 0.0 002 - 0.0 005 0.0 001 0.0 002 - 0.0 003 0.0 001 0.00 06 ~0 0.0 003 ~0 - 0.0 005 0.0 003 - 0.000 2 MCC - 0.0 024 0.0 002 - 0.0 005 0.0 001 0.0 002 - 0.0 003 0.0 001 0.00 06 ~0 0.0 003 ~0 - 0.0 005 0.0 003 - 0.000 2 Decision Tree Confirmation UK 0.0 014 0.0 038 0.0 012 0.0 033 0.0 022 0.0 008 0.0 033 0.00 14 0.0 039 0.0 035 0.0 017 0.0 012 0.0 007 0.000 7 AUC 0.0 014 0.4 64 0.4 704 0.5 013 0.4 88 0.5 159 0.5 379 0.49 1 0.4 964 0.5 235 0.5 007 0.5 168 0.5 077 0.484 Sensitive ness 0.0 014 0.0 038 0.0 012 0.0 033 0.0 022 0.0 008 0.0 033 0.00 14 0.0 039 0.0 035 0.5 007 0.0 012 0.0 007 0.000 7 Definitely k 0.0 014 0.0 038 0.0 012 0.0 033 0.0 022 0.0 008 0.0 033 0.00 14 0.0 039 0.0 035 0.0 017 0.0 012 0.0 007 0.000 7 F1 0.0 014 0.0 038 0.0 012 0.0 033 0.0 022 0.0 008 0.0 033 0.00 14 0.0 039 0.0 035 0.0 017 0.0 012 0.0 007 0.000 7 Kappa - 0.0 024 - 0.0 001 - 0.0 026 - 0.0 005 - 0.0 016 - 0.0 03 - 0.0 005 - 0.00 ~0 - 0.0 003 - 0.0 021 - 0.0 026 - 0.0 031 - 0.003 1 13 MCC - 0.0 024 - 0.0 001 - 0.0 026 - 0.0 005 - 0.0 016 - 0.0 03 - 0.0 005 - 0.00 ~0 - 0.0 003 - 0.0 021 - 0.0 026 - 0.0 031 - 0.003 1 Random Forest (Random Forest) True UK 0.0 019 0.0 041 0.0 013 0.0 03 0.0 027 0.0 008 0.0 043 0.00 0.0 035 0.0 035 0.0 019 0.0 016 0.0 01 0.000 6 AUC 0.5 093 0.4 86 0.5 218 0.4 869 0.4 931 0.5 205 0.5 162 0.47 29 0.5 144 0.4 959 0.5 328 0.5 077 0.4 778 0.498 Sensitive ness 0.0 019 0.0 041 0.0 013 0.0 03 0.0 027 0.0 008 0.0 043 0.00 0.0 035 0.0 035 0.0 019 0.0 016 0.0 01 0.000 6 Definitely k 0.0 019 0.0 041 0.0 013 0.0 03 0.0 027 0.0 008 0.0 043 0.00 0.0 035 0.0 035 0.0 019 0.0 016 0.0 01 0.000 6 F1 0.0 019 0.0 041 0.0 013 0.0 03 0.0 027 0.0 008 0.0 043 0.00 0.0 035 0.0 035 0.0 019 0.0 016 0.0 01 0.000 6 Kappa - 0.0 019 0.0 001 - 0.0 025 - 0.0 008 - 0.0 011 - 0.0 031 0.0 004 - 0.00 23 - 0.0 003 - 0.0 003 - 0.0 019 - 0.0 022 - 0.0 028 - 0.003 2 MCC - 0.0 019 0.0 001 - 0.0 025 - 0.0 008 - 0.0 011 - 0.0 031 0.0 004 - 0.00 23 - 0.0 003 - 0.0 003 - 0.0 019 - 0.0 022 - 0.0 028 - 0.003 2 Radial SVM True UK 0.0 037 0.0 037 0.0 036 0.0 035 0.0 037 0.0 031 0.0 043 0.00 4 0.0 042 0.0 04 0.0 03 0.0 036 0.0 037 0.004 AUC 0.5 092 0.5 126 0.5 13 0.4 958 0.4 791 0.5 0.5 162 0.51 36 0.5 188 0.5 357 0.4 999 0.4 957 0.5 0.484 Sensitive ness 0.0 037 0.0 037 0.0 036 0.0 035 0.0 037 0.0 031 0.0 043 0.00 4 0.0 042 0.0 04 0.0 03 0.0 036 0.0 037 0.004 Definitely k 0.0 037 0.0 037 0.0 036 0.0 035 0.0 037 0.0 031 0.0 043 0.00 4 0.0 042 0.0 04 0.0 03 0.0 036 0.0 037 0.004 F1 0.0 037 0.0 037 0.0 036 0.0 035 0.0 037 0.0 031 0.0 043 0.00 4 0.0 042 0.0 04 0.0 03 0.0 036 0.0 037 0.004 Coat - 0.0 001 - 0.0 002 - 0.0 002 - 0.0 003 - 0.0 001 - 0.0 007 0.0 004 ~0 0.0 003 0.0 002 ~0 - 0.0 002 - 0.0 001 0.000 1 14 MCC - 0.0 001 - 0.0 002 - 0.0 002 - 0.0 003 - 0.0 001 - 0.0 007 0.0 004 ~0 0.0 003 0.0 002 ~0 - 0.0 002 - 0.0 001 0.000 1 Polynomial SVM True UK 0.0 033 0.0 035 0.0 025 0.0 036 0.0 035 0.0 023 0.0 037 0.00 31 0.0 034 0.0 04 0.0 03 0.0 029 0.0 028 0.002 3 AUC 0.4 875 0.5 083 0.5 033 0.4 607 0.4 788 0.4 756 0.4 788 0.48 1 0.4 919 0.5 357 0.4 804 0.5 011 0.4 941 0.478 Sensitive ness 0.0 033 0.0 035 0.0 025 0.0 036 0.0 035 0.0 023 0.0 037 0.00 31 0.0 034 0.0 04 0.0 03 0.0 029 0.0 028 0.002 3 Definitely k 0.0 033 0.0 035 0.0 025 0.0 036 0.0 035 0.0 023 0.0 037 0.00 31 0.0 034 0.0 04 0.0 03 0.0 029 0.0 028 0.002 3 F1 0.0 033 0.0 035 0.0 025 0.0 036 0.0 035 0.0 023 0.0 037 0.00 31 0.0 034 0.0 04 0.0 03 0.0 029 0.0 028 0.002 3 Kappa - 0.0 005 - 0.0 002 - 0.0 013 - 0.0 002 - 0.0 003 - 0.0 015 - 0.0 001 - 0.00 078 - 0.0 004 0.0 002 - 0.0 008 - 0.0 009 - 0.0 01 - 0.001 MCC - 0.0 005 - 0.0 002 - 0.0 013 - 0.0 002 - 0.0 003 - 0.0 015 - 0.0 001 - 0.00 078 - 0.0 004 0.0 002 - 0.0 008 - 0.0 009 - 0.0 01 - 0.001 Linear SVM True UK 0.0 039 1 0.0 04 0.0 033 0.0 037 0.0 039 0.0 035 0.0 039 0.00 4 0.0 04 0.0 038 0.0 035 0.0 034 0.0 038 0.003 AUC 0.5 137 0.5 156 0.4 839 0.5 143 0.4 723 0.4 933 0.4 762 0.51 95 0.5 202 0.5 324 0.5 028 0.4 855 0.4 934 0.500 3 Sensitive ness 0.0 039 1 0.0 04 0.0 033 0.0 037 0.0 039 0.0 035 0.0 039 0.00 4 0.0 04 0.0 038 0.0 035 0.0 034 0.0 038 0.003 Definitely k 0.0 039 1 0.0 04 0.0 033 0.0 037 0.0 039 0.0 035 0.0 039 0.00 4 0.0 04 0.0 038 0.0 035 0.0 034 0.0 038 0.003 F1 0.0 039 0.0 04 0.0 033 0.0 037 0.0 039 0.0 035 0.0 039 0.00 4 0.0 04 0.0 038 0.0 035 0.0 034 0.0 038 0.003 1 Kappa ~0 ~0 - 0.0 005 ~0 ~0 - 0.0 003 ~0 0.00 01 0.0 001 ~0 - 0.0 003 - 0.0 005 ~0 - 0.000 3 MCC ~0 ~0 - 0.0 005 ~0 ~0 - 0.0 003 ~0 0.00 01 0.0 001 ~0 - 0.0 003 - 0.0 005 ~0 - 0.000 3 K-means True UK 0.0 039 0.0 044 0.0 036 0.0 036 0.0 037 0.0 04 0.0 041 0.00 35 0.0 042 0.0 036 0.0 037 0.0 046 0.0 036 0.004 AUC 0.4 865 0.5 12 0.5 022 0.5 154 0.4 894 0.5 039 0.5 023 0.48 29 0.5 002 0.5 303 0.5 098 0.4 928 0.4 984 0.51 Sensitive ness 0.0 039 1 0.0 044 0.0 036 0.0 036 0.0 037 0.0 04 0.0 041 0.00 35 0.0 042 0.0 036 0.0 037 0.0 046 0.0 036 0.004 Definitely k 0.0 039 0.0 044 0.0 036 0.0 036 0.0 037 0.0 04 0.0 041 0.00 35 0.0 042 0.0 036 0.0 037 0.0 046 0.0 036 0.004 F1 0.0 039 0.0 044 0.0 036 0.0 036 0.0 037 0.0 04 0.0 041 0.00 35 0.0 042 0.0 036 0.0 037 0.0 046 0.0 036 0.004 Kappa 0.0 002 0.0 002 ~0 - 0.0 001 ~0 0.0 001 0.0 002 0.00 038 0.0 003 - 0.0 003 - 0.0 001 0.0 004 - 0.0 001 ~0 MCC 0.0 002 0.0 003 ~0 - 0.0 001 ~0 0.0 001 0.0 002 0.00 038 0.0 003 - 0.0 003 - 0.0 001 0.0 004 - 0.0 001 ~0 Neural Network Confirmation UK 0.0 034 0.0 04 0.0 038 0.0 038 0.0 038 0.0 039 0.0 04 0.00 39 0.0 042 0.0 041 0.0 041 0.0 037 0.0 036 0.004 AUC 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 Sensitive ness 0.0 039 0.0 04 0.0 038 0.0 038 0.0 038 0.0 039 0.0 04 0.00 39 0.0 042 0.0 041 0.0 041 0.0 037 0.0 036 0.004 Definitely k 0.0 034 0.0 04 0.0 038 0.0 038 0.0 038 0.0 039 0.0 04 0.00 39 0.0 042 0.0 041 0.0 041 0.0 037 0.0 036 0.004 16 F1 0.0 034 0.0 04 0.0 038 0.0 038 0.0 038 0.0 039 0.0 04 0.00 39 0.0 042 0.0 041 0.0 041 0.0 037 0.0 036 0.004 Kappa ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 MCC ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 LSTM Verified UK 0.0 034 0.0 04 0.0 038 0.0 038 0.0 038 0.0 039 0.0 04 0.00 39 0.0 042 0.0 041 0.0 041 0.0 037 0.0 036 0.004 AUC 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 Sensitive ness 0.9 965 0.9 952 0.9 965 0.9 959 0.9 96 0.9 959 0.9 959 0.99 59 0.9 957 0.9 958 0.9 962 0.9 957 0.9 959 0.996 3 Definitely k 0.9 965 0.9 952 0.9 965 0.9 959 0.9 96 0.9 959 0.9 959 0.99 59 0.9 957 0.9 958 0.9 962 0.9 957 0.9 959 0.996 3 F1 0.9 965 0.9 952 0.9 965 0.9 959 0.9 96 0.9 959 0.9 959 0.99 59 0.9 957 0.9 958 0.9 962 0.9 957 0.9 959 0.996 3 Kappa ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 MCC ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 ~0 Table 3. ML (Machine Learning) Applied to the Invention Consequences of the Attacks 17 References: [1] Nolting, S. (2023). The neoTRNG V2- A Tiny and Platform-Independent True Random Number Generator for any FPGA. (Version v3). Zenodo. https: / / doi.org / 10.5281 / zenodo.10071201 [2] Valeria Piscopo. 2024. Design of a True Random Number Generator for Post-quantum Cryptography. Ph. D. Dissertation. Politecnico di Torino. [3] Rafael Oliveira, João Cabacinho, Luís Oliveira, and João Casaleiro. 2023. TrueRandom Number Generator Prototype Implemented in an FPGA. In 2023 7th International Young 10 Engineers Forum (YEF-ECE). 95–99. https: / / doi.org / 10.1109 / YEF- ECE58420.2023.10209297 [4] Fabio Frustaci, Fanny Spagnolo, Stefania Perri, and Pasquale Corsonello. 2023. A High- Speed FPGA-Based True Random Number Generator Using Metastability With Clock 15 Managers. IEEE Transactions on Circuits and Systems II: Express Briefs 70, 2 (2023), 756–760. https: / / doi.org / 10.1109 / TCSII.2022.3211278 [5] Junjun Wang, Zhao Huang, Lirong Zhou, Jinhui Liu, Xiaohong Jiang, Yin Chen, Yuan Cao, and Quan Wang. 2024. A New Entropy Source Design Based on NAND-XOR Ring 20 Oscillators for Resource-Efficient and Ultra-High Throughput TRNG. IEICE Electronics Express advpub (2024), 21.20240259. https: / / doi.org / 10.1587 / elex.21.20240259 [6] Miloš Grujić and Ingrid Verbauwhede. 2022. TROT: A Three-Edge Ring Oscillator Based True Random Number Generator With Time-to-Digital Conversion. IEEE Transactions on 25 Circuits and Systems I: Regular Papers 69, 6 (2022), 2435–2448. https: / / doi.org / 10.1109 / TCSI.2022.3158022

Claims

18 REQUESTS 1.5  at least one entropy source used as an entropy source for the random number generator cell (1),  De-Biasing unit (2) which detects randomness in the bit series given as input,  Collects the bits extracted by the De-Biasing unit (2) and is byte-wide output giving sample unit (3) 10 including, key and unique in various applications in cryptography and secure communication. DDR-based low-resource and low-power applications used for the production of disposable ribbons. with high consumption and high disorganization in cryptographic and security units, a truly random number generator resistant to machine learning attacks It is a method, and its characteristic is; 15  Entropy cell (1), De-Biasing unit (2), sampling unit (3) integrated The system is activated and random number generation begins. (1001),  data of entropy sources located within the mentioned entropy cell (1) start generating entropy-input data (1002), 20  If the entropy cell (1) is a ring oscillator, by its very nature it has its own structure. if it generates 1 bit for itself, and the entropy source is DDR. External inputs are obtained from RAM and the generated 3-bit value is used. Initiating the XOR operation (1003),  The value of the 1-bit value which is the output of the XOR operation is assigned to the De-Biasing unit (2) 25 given as input (1004),  XORed data from the entropy cell (1) of the De-Biasing unit (2) According to the system, it follows the state of randomness (1005),  In the De-Biasing unit (2), the bits that have received confirmation for randomness are sent to the sampling unit. (3) and stored until 1 byte of data is collected and 1 byte is 30 when a new random number is given as output (1006) It includes the steps involved in the process. 35

Images