Random number generator for providing authentic random numbers

Abstract

The invention relates to a random number generator (1) for providing authentic random numbers for a probabilistic computer (2), comprising a chaotic light source (4), which is designed to continuously emit light of random intensities according to a distribution having an average value, at least one intensity modulator (6), wherein the intensity modulator (6) is optically connected to the chaotic light source (4) for transmitting the light emitted by the chaotic light source (4), the intensity modulator (6) is designed to modulate the average value of the distribution of the intensities with an electro-optical pulse, to encode same in a symbol (8) and to output same to at least one optical filter (10) which is optically connected to the intensity modulator (6) for transmission, wherein the optical filter (10) is selectively permeable for an optical bandwidth of the symbol (8), the optical filter (10) is optically connected to at least one detector (12) for transmission of the symbol (8), and the detector (12) is designed to measure the intensity of the light of the optical bandwidth, and an authentic random number is provided by measuring an intensity from the modulated distribution of the symbol (8). In this way, a faster and more scalable random number generator (1) is provided.

Description

[0001] Random number generator for providing truly random numbers

[0002] The invention relates to a random number generator for providing true random numbers for a probabilistic computer, comprising a chaotic light source which is designed to continuously emit light of random intensities.

[0003] In general, deterministic computations are well-suited for passing data through a known model. In contrast, the derivation of model parameters from incomplete observational data is an inherently probabilistic process. Inference using deterministic artificial neural networks on deterministic hardware leads to astonishing results in areas such as medical diagnostics, autonomous driving, and natural language models. However, such deterministic models are point estimates based on known data and do not consider underlying complete statistical distributions.

[0004] Processing complex probabilistic models on conventional deterministic hardware therefore presents challenges. Integrals describing the probabilistic model are unsolvable even for a small number of parameters and must be approximated by sampling the distributions. Therefore, high-speed random number generators that provide or sample truly random numbers are needed, along with an architecture capable of evaluating the model for all drawn samples in a reasonable time. This architecture then roughly corresponds to a probabilistic computer that feeds the random numbers generated by the random number generator to a probabilistic processor.

[0005] Various approaches are conceivable for generating these truly random numbers. In physical computing, for example, the inherent randomness of physical processes is exploited to implement probabilistic bits by using stochastic magnetic tunnel junctions or fluctuations in optical parametric oscillators. Similarly, the randomness of phase transitions in phase-change material (PCM) cells can be used for probabilistic modeling, although it is limited by the PCM's limited number of switching cycles. However, some of these approaches are poorly scalable, very complex, and / or rely on random structural changes in materials, making them only partially suitable for high-speed computing or the associated need to generate random numbers at a specific speed.

[0006] Based on this, the object of the present invention is to provide a fast and scalable random number generator for true random numbers.

[0007] This object is achieved by the subject matter of the independent claims. Preferred developments of the invention are described in the subclaims.

[0008] According to the invention, a random number generator for providing true random numbers for a probabilistic computer is thus provided, comprising a chaotic light source which is designed to continuously emit light of random intensities according to a distribution with a mean value, at least one intensity modulator, wherein the intensity modulator is optically connected to the chaotic light source for transmitting the light emitted by the chaotic light source, the intensity modulator is designed to modulate the mean value of the distribution of the intensities with an electro-optical pulse and to encode intensities of such a modulated distribution with a modulated mean value in a symbol and to output them to at least one optical filter optically connected to the intensity modulator for transmission, wherein the optical filter is selectively permeable to an optical bandwidth of the symbol,the optical filter is optically connected to at least one detector for transmitting the symbol, and the detector is designed to measure the intensity of the light of the optical bandwidth and by measuring an intensity from the modulated distribution of the symbol, a true random number is provided.

[0009] A chaotic light source is understood here as a light source that emits chaotic light generated by thermal radiation or amplified spontaneous emission. The chaotic light source serves as the entropy source of the random number generator and continuously emits broadband light with fluctuating intensities in the form of noise. This noise is proportional to the intensity of the light emitted by the chaotic light source and can therefore be changed by intensity modulation. The fluctuating intensities of the noise of the chaotic light source follow a statistical distribution when integrated. The intensities are therefore emitted or sampled according to this distribution. The chaotic light source is therefore used to sample or draw the intensities of the emitted light from this distribution.By modulating the mean, i.e., modulating the distribution with the intensity modulator, the distribution can then be adjusted. Modulating the mean implies changing or modulating the distribution toward the modulated distribution with this modulated mean. Modulating thus modulates the distribution to exhibit the modulated mean, which is a property of the distribution.

[0010] According to the invention, the electro-optical pulse of a specific pulse shape is modulated onto the intensity distribution using the intensity modulator. Essentially, the mean value is thereby reduced using the intensity modulator. The intensity modulator serves, in this case, to attenuate the intensities of the emitted light. In this way, it is possible to generate true random numbers at intervals of a few picoseconds. The intensities emitted from the chaotic light source follow a specific underlying distribution and are then modulated in the intensity modulator using the electro-optical pulse, so that the intensities from the modulated distribution are encoded into the symbol and output. This encoded symbol or symbol comprises the modulated mean value of the modulated distribution, which in turn is only encoded in the waveform.The intensity modulator can be an electro-absorption modulator or an electro-optical modulator. The electro-optical modulator is particularly suitable because it allows for high contrast.

[0011] The intensities of the symbol measured by the detector are then proportional to voltages that are output by the detector and can in turn be used. An optical connection is understood here to be a connection that is able to transmit the light emitted by the chaotic light source or, subsequently, the symbol, i.e. the light emitted after the modulated distribution. For this purpose, waveguides and fiber or fiber optic connections can be provided, for example, which are suitable for transmitting light with a corresponding optical bandwidth. However, it can also be provided that the respective components, such as the chaotic light source, the intensity modulator, the optical filter and the detector, are arranged next to one another in such a way that the respective outputs of the components are directly adjacent to the respective inputs of the components and the light orthe symbol is subsequently transmitted without fiber optic connections.

[0012] The electro-optical pulses used by the intensity modulator to modulate the distribution can be transmitted to the intensity modulator via an electrical input interface, ultimately outputting corresponding symbols. An electrical output interface can then be provided directly at the detector, with which the true random numbers can be assigned to the respective symbol. The intensity modulator and the detector are thus connected to each other via the electrical input interface and the electrical output interface in the form of an electrical system. The electrical system allows the respective electro-optical pulse shapes as well as the time intervals between the pulse shapes, a clocking, to be adjusted.

[0013] In principle, it is possible to design the random number generator in various ways. According to a preferred development of the invention, however, the random number generator is provided with a plurality of intensity modulators, each intensity modulator being connected to the chaotic light source by means of an optical delay line for transmitting the emitted light, the delay lines being configured to allow different transmission times between the individual delay lines, and each intensity modulator being used to encode a symbol. Using the delay lines, it is possible to generate a plurality of mutually uncorrelated chaotic light sources for the respective intensity modulator from a single chaotic light source. This can be achieved, for example, by means of length and / or material differences in the delay lines.Alternatively, several chaotic light sources can each be assigned to an intensity modulator. However, transmitting the emitted light through a delay line is far more efficient and easier to scale. Each intensity modulator is then assigned at least one optical filter and detector optically connected to the intensity modulator. Thus, uncorrelated true random numbers are generated corresponding to the number of intensity modulators.

[0014] In principle, it is possible to operate the random number generator with an optical filter. However, according to a preferred development of the invention, the random number generator is provided with a plurality of optical filters optically connected to the intensity modulator, such that the symbol can be transmitted to each optical filter, and with a plurality of detectors, wherein the optical filters filter different optical bandwidths of the symbol, wherein one detector is optically connected to one optical filter each, and the detectors are designed to measure the intensity of the light in the different optical bandwidths, such that by measuring intensities from the modulated distribution of the symbol, a plurality of true random numbers are provided. The plurality of optical filters thus forms a demultiplexer, with which inverse demultiplexing in the wavelength can be carried out.This demultiplexing makes it possible to simultaneously obtain a number of true random numbers corresponding to the number of detectors and optical filters. The light emitted by the chaotic light source is broadband and can have frequencies within an optical bandwidth of, for example, 1 to 10 THz. Thus, if each optical filter has an optical bandwidth of 100 GHz and the detector is correspondingly sensitive to the respective optical bandwidth, 10 true random numbers can be provided simultaneously with an optical bandwidth of 1 THz. Since different optical bandwidths of the light from the chaotic light source are measured, the distributions are uncorrelated. Thus, true random numbers corresponding to the number of detectors and uncorrelated to one another are generated simultaneously.If, in this case, a plurality of intensity modulators with corresponding delay lines are additionally used, the number of provided uncorrelated true random numbers can be exponentiated.

[0015] It is possible to modulate the distribution in various ways. However, according to a preferred development of the invention, the intensity modulator is configured to modulate the mean value of the distribution of random intensities with a plurality of electro-optical pulses and to encode the intensities of such modulated distributions with modulated mean values ​​in the form of a plurality of symbols. The detector is configured to sum a predetermined number of the plurality of symbols, so that by measuring an intensity from a summed distribution of the summed symbols, a true random number is provided.

[0016] In this context, according to a particularly preferred development of the invention, it is provided that the intensity modulator is designed to successively modulate the mean value of the distribution of the intensities with the plurality of electro-optical pulses.

[0017] The random intensities are continuously emitted by the chaotic light source. This makes it possible to encode different mean values ​​in the form of symbols for the continuous stream of intensities using the intensity modulator, or to sample the intensities from the distribution using the mean values ​​modulated accordingly using the intensity modulator. If these symbols are then summed using the detector, the mean of the summed distribution corresponds to the sum of the means of the distributions, i.e. the sum of the predetermined number of symbols. By summing the individual distributions of the symbols, the ratio of the mean to a standard deviation of the distribution can also be varied, making the distribution narrower or broader. If long, flat pulse shapes are modulated with the intensity modulator, a narrow summed distribution with a lower standard deviation is obtained.If short, high-intensity pulse shapes are modulated with the intensity modulator, a broad summed distribution with a larger standard deviation is obtained. Successive modulation allows for a particularly advantageous adjustment of the summed distribution, which naturally requires a longer exposure time in the detector to calculate the sum over the predetermined number of symbols.

[0018] In principle, it is possible to form the sum of symbols over different numbers of symbols. However, according to a preferred development of the invention, the predetermined number of symbols corresponds to nine symbols. Coding the mean value in a single large symbol causes more noise than coding in nine smaller symbols. This then also allows the shape of the modulated distribution to be designed particularly advantageously. Modulated distributions with smaller standard deviations are achieved because the noise partially cancels out. In this way, the mean value and the standard deviation of the output distribution can be adjusted independently of one another. An advantage of this coding scheme is that the readout circuit always performs the same operations, i.e. summing over nine symbols, and requires no information about the noise distribution.

[0019] In principle, it is possible to use different chaotic light sources with respective different distributions. According to a preferred embodiment of the invention, however, the chaotic light source is an amplified spontaneous emission source. This continuously emits light of random intensities based on amplified spontaneous emission. According to a particularly preferred embodiment of the invention, the distribution according to which the light of random intensities is emitted is a Bose-Einstein distribution. This means that the fluctuating intensities of the noise of the chaotic light source, when integrated, correspond to a Bose-Einstein distribution. The intensities are therefore emitted or sampled according to the Bose-Einstein distribution. One form of the Bose-Einstein distribution is influenced by a degeneracy factor M, so that an M-fold Bose-Einstein distribution is referred to.

[0020] The detector and the electrical system have a certain electrical bandwidth, which corresponds to a low-pass filter. This degeneracy factor can be adjusted and the distribution modulated by a ratio of the optical bandwidth of the optical filter to the electrical bandwidth of the detector and the electrical system. A degeneracy factor of M = 10 is particularly preferred in this context. The electrical bandwidth is 10 GHz, so with an optical bandwidth of 100 GHz, a degeneracy factor of M = 10 is achieved.

[0021] The invention further relates to a probabilistic computer comprising a random number generator as described above and a probabilistic processor, wherein the probabilistic processor has a plurality of intensity modulators, each of which is optically connected to a processor input for transmitting the respective symbol, and at least one processor output of the probabilistic processor is optically connected to at least one optical filter. This makes it possible to provide the probabilistic processor with the various modulated distributions at the processor inputs and to allow these to communicate with one another in the probabilistic processor. The probabilistic processor is a non-deterministic, inherently probabilistic processor, which according to a preferred embodiment of the invention is a photonic crossbar array.The use of such a probabilistic processor together with the random number generator enables the construction of a purely optical probabilistic computer. This, in turn, enables the efficient use of true random numbers in conjunction with a high baud rate. A separate random number generator and its interfaces can also be avoided.

[0022] The random numbers are generated optically. The electrical system is deterministic and provides the properties of the distributions to the probabilistic processor in a quasi-deterministic manner. However, the probabilistic processor itself draws the random numbers according to the distributions. The random numbers are thus generated "automatically" in the processor. Only the distribution needs to be specified accordingly.

[0023] The photonic crossbar array is an arrangement of optical switching elements arranged in a grid or matrix. These switching elements enable the targeted switching or routing of optical signals from one optical line to another. The photonic crossbar array can be used in optical networks, switching systems, or other optical communication applications.

[0024] In principle, it is possible to use different probabilistic processors that can fulfill different functions. However, according to a preferred embodiment of the invention, the probabilistic processor is designed for matrix-vector multiplication. In this context, according to a further preferred embodiment of the invention, it is particularly preferred that the probabilistic processor be configured to form an artificial neural network.

[0025] For artificial neural networks, such as Bayesian neural networks, probabilistic processors are particularly advantageous, as they can perform matrix-vector multiplication. An activation layer of the artificial neural network is defined by the processor inputs, to which the symbols corresponding to activation distributions are applied. Matrix propagation in the probabilistic processor then represents a further layer. The process of generating truly random numbers is thus directly linked to optical computing in the probabilistic processor.

[0026] In principle, it is possible to construct the probabilistic computer in various ways. However, according to a preferred embodiment of the invention, the random number generator and the probabilistic processor are embodied in the form of a single integrated circuit. This enables a significantly more compact design, so that, apart from the delay lines, no fiber optic connections are required for light transmission. For example, the light can be transmitted directly from an output of the intensity modulator to a processor input.

[0027] The invention is described in more detail below with reference to the drawings using preferred embodiments.

[0028] In the drawing show

[0029] Fig. 1 schematically shows a random number generator for providing real

[0030] Random numbers for a probabilistic computer according to a preferred embodiment of the invention,

[0031] Fig. 2 schematically shows a random number generator for providing real

[0032] Random numbers for a probabilistic computer according to another preferred embodiment of the invention and

[0033] Fig. 3 shows a probabilistic computer with the random number generator according to an embodiment of the invention.

[0034] Fig. 1 schematically shows a random number generator 1 for providing truly random numbers for a probabilistic computer 2 according to a preferred embodiment of the invention. A chaotic light source 4 based on amplified spontaneous emission is used as the entropy source. The amplified spontaneous emission is generated by an erbium-doped fiber and can be modeled as a superposition of many emitters with different wavelengths, each having a random phase and amplitude. Hovering between different frequency components leads to a temporally fluctuating optical intensity. Overall, the integration of the fluctuating intensities over a fixed time interval yields an M-fold Bose-Einstein distribution of the integrated intensities. The degeneracy factor M depends on the number of independent temporal coherence cells of the chaotic light source 4 within the time interval.The temporal coherence time is with an optical bandwidth v. opl of the chaotic light source by the Wiener-Khinchin theorem, and the time interval is related to the electrical bandwidth v el of an electrical system. Thus, v opl and v el the global parameters that define the probabilistic properties of the random number generator 1.

[0035] The electrical system has an electrical input interface connected to an electrical output interface. The electrical input interface is also connected to an intensity modulator 6, which in this case is an electro-optical modulator 6. The electrical output interface is connected to a detector 12. The electro-optical modulator 6 is connected to the chaotic light source 4 via fiber optics. An optical filter 10 is connected via fiber optics between the detector 12 and the electro-optical modulator 6.

[0036] Nine electro-optical pulses are successively modulated by the electro-optical modulator 6 onto the distribution, which is transmitted via fiber optics from the chaotic light source 4 to the electro-optical modulator 6. These resulting pulse shapes, the modulated distributions, are successively encoded into nine symbols 8 by the electro-optical modulator 6. One symbol length of a symbol corresponds to one coherence cell. For each individual symbol 8, the mean value of the intensity distribution was modulated with the respective electro-optical pulse, and the respective intensities of such a modulated distribution were encoded into the respective symbol 8 with a modulated mean value. The symbols 8 in Fig. 1 indicate the respective modulated mean value of the modulated distribution. These symbols 8 are finally selectively filtered in the optical filter 10 so that only light of a specific optical bandwidth can pass through the optical filter 10.The nine symbols are successively summed by detector 12. The intensity measured by detector 12 from the sum of the nine symbols 8 is then a true random number, the summed distribution.

[0037] Fig. 2 schematically shows the random number generator 1 for providing true random numbers for the probabilistic computer 2 with multiple detectors 12. The light from the chaotic light source 4 is broadband and is filtered by multiple optical filters 10 such that each detector 12 receives an optical bandwidth of the light. The broadband light is thus demultiplexed by the optical filters 10. The modulated distribution in the symbol 8 is identical for each optical filter 10 and each detector 12. Since the respective intensities of the different bandwidths are uncorrelated, several uncorrelated true random numbers can be sampled. The modulated distributions are sampled in parallel.

[0038] Finally, Fig. 3 shows the probabilistic computer 2 with the random number generator 1 and a probabilistic processor 16. The light from the chaotic light source 4 is split and connected via four delay lines 14, each connected to an electro-optical modulator 6. This virtually generates four uncorrelated distributions, which are modulated accordingly in the electro-optical modulators 6 or encoded into the symbols 8. The symbols 8 have a symbol length of 56.8 ps. The electro-optical modulators 6 are operated at 17.6 GBaud. The electrical system has an electrical bandwidth of 30 GHz. Each of the four electro-optical modulators 6 is connected via fiber optics to a processor input 18 of the probabilistic processor, which in this case is a photonic crossbar array.The photonic crossbar array is designed for matrix-vector multiplications and performs multiplication and addition operations on the transmitted symbols 8 based on the non-volatile phase-change material germanium-antimony-tellurium. Additions are performed by overlapping the symbols 8 in a waveguide. Multiplications are performed by attenuating the symbols 8 in germanium-antimony-tellurium cells. The germanium-antimony-tellurium cells are applied to the respective waveguides in the probabilistic processor 16 and can be switched between a highly absorbing crystalline state and a barely absorbing amorphous state. Both states are non-volatile, enabling in-memory computing without the need for a constant power supply to maintain the respective state.Both addition and multiplication are phase-insensitive and therefore compatible with chaotic light sources 4 that emit random spectral phases.

[0039] In the present embodiment, the probabilistic processor 16 has four processor outputs 20. Each processor output 20 is connected to four optical filters 10 for transmitting the symbols 8. The optical filters 10 are used to demultiplex at each processor output 20 according to the 200 GHz ITU grid, so that the intensities of the wavelength channels C28, C30, C31, and C34 are measured at the detectors 12 connected to the optical filters 10.

[0040] There are two main contributions to the modulated distribution for the modulated mean x of symbol 8. First, there are fluctuations in the intensities due to the light emitted by the chaotic light source 4, which is described by an M-fold Bose-Einstein distribution. Therefore, there are fluctuations in the intensity measured by detector 12 that are proportional to the modulated mean. Due to a high photon number, optical shot noise is negligible. Second, there is an electronic noise floor that can be described by a Gaussian distribution with standard deviation λ. In an idealized system, the modulated distribution is the convolution of these two independent random processes:

[0041] Since the modulated mean depends on the symbol 8, it is possible to shape the modulated distribution by changing the electro-optical pulse shape. With a modulated mean of zero, electrical noise is the dominant random source, resulting in a Gaussian shape of the output distribution. As the modulated mean increases, the fluctuations in the intensities, described by an M-fold Bose-Einstein distribution, become the distribution.

[0042] The invention has received funding from the European Union's Horizon 2020 research and innovation program under the number 101017237.

[0043] List of reference symbols

[0044] 1 random number generator

[0045] 2 probabilistic computer 4 chaotic light source

[0046] 6 Intensity modulator

[0047] 8 Symbol

[0048] 10 optical filters

[0049] 12 Detector 14 Delay line

[0050] 16 probabilistic processor

[0051] 18 Processor input

[0052] 20 Processor output

Claims

Patent claims 1. Random number generator (1) for providing genuine random numbers for a probabilistic computer (2), comprising a chaotic light source (4) designed to continuously emit light of random intensities according to a distribution with a mean value, at least one intensity modulator (6), wherein the intensity modulator (6) is optically connected to the chaotic light source (4) for transmitting the light emitted by the chaotic light source (4), the intensity modulator (6) is designed to modulate the mean value of the distribution of intensities with an electro-optical pulse and to encode intensities of such a modulated distribution with a modulated mean value in a symbol (8) and to output them to at least one optical filter (10) optically connected to the intensity modulator (6) for transmission, wherein the optical filter (10) is selectively permeable to an optical bandwidth of the symbol (8),the optical filter (10) is optically connected to at least one detector (12) for transmitting the symbol (8), and the detector (12) is designed to measure the intensity of the light of the optical bandwidth and by measuring an intensity from the modulated distribution of the symbol (8), a true random number is provided.

2. Random number generator (1) according to claim 1, comprising a plurality of intensity modulators (6), wherein each intensity modulator (6) is connected to the chaotic light source (4) by means of an optical delay line (14) for transmitting the emitted light, wherein the delay lines (14) are configured to allow different transmission times between the individual delay lines (14), and wherein each intensity modulator (6) encodes one symbol (8).

3. Random number generator (1) according to one of claims 1 or 2 with a plurality of optical filters (10) optically connected to the intensity modulator (6), so that the Symbol (8) is transmittable to each optical filter (10), and with a plurality of detectors (12), wherein the optical filters (10) filter different optical bandwidths of the symbol (8), wherein each detector (12) is optically connected to each optical filter (10), and the detectors (12) are designed to measure the intensity of the light in the different optical bandwidths, so that by measuring intensities from the modulated distribution of the symbol (8) a plurality of true random numbers are provided.

4. Random number generator (1) according to one of the preceding claims, wherein the intensity modulator (6) is designed to modulate the mean value of the distribution of random intensities with a plurality of electro-optical pulses and to encode the intensities of such modulated distributions with modulated mean values in the form of a plurality of symbols (8), and the detector (12) is designed to sum a predetermined number from the plurality of symbols (8), so that by measuring an intensity from a summed distribution of the summed symbols (8) a true random number is provided.

5. Random number generator (1) according to claim 4, wherein the intensity modulator (6) is designed to successively modulate the mean value of the distribution of the intensities with the plurality of electro-optical pulses.

6. Random number generator (1) according to one of claims 4 or 5, wherein the predetermined number of symbols (8) corresponds to nine symbols (8).

7. Random number generator (1) according to one of the preceding claims, wherein the chaotic light source (4) is an amplified spontaneous emission source.

8. Random number generator (1) according to one of the preceding claims, wherein the distribution according to which the light of random intensities is emitted is a Bose-Einstein distribution.

9. Probabilistic computer (2) comprising a random number generator (1) according to one of claims 2 to 8 and a probabilistic processor (16), wherein the probabilistic processor (16) has a plurality of processor inputs (18) and an intensity modulator (6) is optically connected to a respective processor input (18) for transmitting the respective symbol (8), and at least one processor output (20) of the probabilistic processor is optically connected to at least one optical filter (10).

10. Probabilistic computer (2) according to claim 9, wherein the probabilistic processor (16) is a photonic crossbar array.

11. Probabilistic computer (2) according to one of claims 9 or 10, wherein the probabilistic processor (16) is designed for matrix-vector multiplication.

12. Probabilistic computer (2) according to claim 11, wherein an artificial neural network is formed with the probabilistic processor (16).

13. Probabilistic computer (2) according to one of claims 9 to 12, wherein the random number generator (1) and the probabilistic processor (16) are designed in the form of a single integrated circuit.

Images