A PBIT generator and probability computation system based on random triangular waves

By using a PBIT generator based on random triangular waves, highly independent and high-precision probability bits are generated, solving the problems of non-independent and highly correlated probability bits in existing technologies, and improving computational accuracy and stability.

CN123114416APending Publication Date: 2026-11-17SUZHOU INSTON TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611087306.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-21
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing PBIT generators are insufficient in terms of high precision controllability, resistance to electromagnetic interference, and stable repeatability, resulting in non-independent probability bit outputs and strong correlations, which affect computational accuracy and stability.

Method used

A PBIT generator based on random triangular waves is adopted. A random triangular wave signal is generated by a random signal generation module, a control threshold signal is set by a threshold configuration module, and sampling comparison is performed by a dynamic comparison module to output probability bits with high independence and high precision.

Benefits of technology

It achieves statistically independent and uncorrelated probability bit outputs, improving the accuracy and stability of calculations and reducing hardware power consumption.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123114416A_ABST
    Figure CN123114416A_ABST
Patent Text Reader

Abstract

This application provides a PBIT generator and probability calculation system based on random triangular waves. The PBIT generator provided in this application includes a random signal generation module, a threshold configuration module, and a dynamic comparison module. The random signal generation module generates a random triangular wave signal with random slope characteristics based on a random data stream. The threshold configuration module provides a control threshold signal for setting a target probability resolution. The dynamic comparison module is connected to both the random signal generation module and the threshold configuration module, and, triggered by a clock signal, samples and compares the random triangular wave signal with the control threshold signal, outputting an initial probability bit corresponding to the desired flip probability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of digital communication technology, and in particular to a PBIT generator and probability calculation system based on random triangular waves. Background Technology

[0002] With the development of artificial intelligence and uncertain computing, traditional architectures face severe computing power bottlenecks when dealing with complex problems such as combinatorial optimization. As the core underlying unit of probabilistic computing, the probability bit generator (PBIT) can randomly flip between logic 0 and 1 with a specific probability. It is the hardware cornerstone for breaking through this bottleneck and building high-performance probabilistic computing chips.

[0003] Currently, PBIT generation is mainly achieved through two schemes: one is to capture the physical entropy source by using nanodevices such as magnetic tunnel junctions (MTJs) to capture random magnetic moment reversals under thermal fluctuations; the other is to use a traditional pseudo-random number generator based on standard CMOS technology to generate random waveforms and then map the target probability through numerical comparison. However, the probability output function of the MTJ-based scheme has highly nonlinear characteristics and is extremely sensitive to operating temperature, drain voltage, and external stray magnetic fields, making it extremely difficult to achieve high precision control and stable repeatability. On the other hand, the scheme based on CMOS technology to generate analog noise or a single pseudo-random waveform is prone to having periodic residues or spatial coupling patterns that cannot be completely eliminated due to the inherent deterministic logic of the circuit. This leads to calculation distortion due to strong correlation when multiple signals are multiplied or added in subsequent stages.

[0004] Therefore, there is an urgent need for a PBIT generator based on random triangular waves, which can output statistically independent and uncorrelated probability bits by establishing a highly precise, controllable, electromagnetically interference-resistant, and stably repeatable deterministic mapping relationship. Summary of the Invention

[0005] In view of this, this application provides a PBIT generator and probability operation system based on random triangular waves, which can output statistically independent and uncorrelated probability bits by establishing a highly precise, controllable, electromagnetic interference resistant and stable repeatable deterministic mapping relationship.

[0006] Specifically, this application is implemented through the following technical solution:

[0007] The first aspect of this application provides a PBIT generator based on random triangular waves, the PBIT generator including a random signal generation module, a threshold configuration module and a dynamic comparison module;

[0008] The random signal generation module is used to generate a random triangular wave signal with random slope characteristics based on the random data stream.

[0009] The threshold configuration module is used to provide a control threshold signal for setting the target probability resolution;

[0010] The dynamic comparison module is connected to the random signal generation module and the threshold configuration module respectively. It is used to sample and compare the random triangular wave signal with the control threshold signal under the trigger of the clock signal, and output the initial probability bits corresponding to the expected flip probability.

[0011] Optionally, the random signal generation module includes: multiple independent random symbol generation sub-modules for generating multiple uncorrelated analog symbol signals in parallel;

[0012] A channel selector, wherein multiple input terminals of the channel selector are respectively connected to the output terminals of each of the random symbol generation submodules;

[0013] And a data selection controller, connected to the control terminal of the channel selector, is used to control the channel selector to dynamically and randomly switch the output of the multiple analog symbol signals according to the input control random number, and combine them to form a random triangular wave signal.

[0014] Optionally, the triangular wave symbol oscillates periodically between a first amplitude threshold and a second amplitude threshold, and the slope feature is characterized by the transit time of the triangular wave symbol between the first amplitude threshold and the second amplitude threshold.

[0015] The rise transit time of the triangular wave symbol from the first amplitude threshold to the second amplitude threshold is selectively limited to a first time threshold interval or a second time threshold interval according to the random number bit stream.

[0016] The descent transit time of the triangular wave symbol from the second amplitude threshold to the first amplitude threshold is selectively limited to a third time threshold interval or a fourth time threshold interval according to the random number bit stream;

[0017] The maximum value of the first time threshold interval is less than the minimum value of the second time threshold interval, and the maximum value of the third time threshold interval is less than the minimum value of the fourth time threshold interval.

[0018] Optionally, the signal generation rate of the random number generator is greater than twice the symbol transformation characteristic frequency of the symbol converter.

[0019] Optionally, the threshold configuration module includes a digital-to-analog converter.

[0020] Optionally, the digital-to-analog converter is specifically a voltage divider resistor network used to output multiple sets of different control threshold signals, which are then output by the dynamic comparison module as corresponding multi-path constant probability bits. These multi-path constant probability bits are combined to form the initial probability bits.

[0021] Optionally, the dynamic comparison module employs a dynamic comparator.

[0022] The second aspect of this application provides a probability calculation system, which includes a plurality of PBIT generators based on random triangular waves as described in any one of the claims provided in the first aspect of this application, and a probability calculation circuit.

[0023] The random sources used by each of the PBIT generators are independent of each other, and the initial probability bits output by each of the PBIT generators are independent and uncorrelated.

[0024] The probability calculation circuit is connected to the output terminal of each of the PBIT generators and is used to perform combination calculations on the initial probability bits output by each of the PBIT generators to obtain the target probability bits; the precision of the target probability bits is higher than that of the initial probability bits output by each of the PBIT generators.

[0025] Optionally, the probability operation circuit uses an AND gate as a probability multiplier and a parallel counter as a probability adder.

[0026] The PBIT generator and probability calculation system based on random triangular waves provided in this application firstly, at the system source, a random signal generation module generates a random triangular wave signal with random slope characteristics based on a random data stream, producing an analog waveform with irregular abrupt changes in voltage amplitude and time axis. This eliminates the periodic residuals and harmonic coherence of traditional triangular waves from a physical mechanism, ensuring the randomness of the system's underlying structure. Next, a threshold configuration module provides a control threshold signal for setting the target probability resolution, converting the probability precision in the digital domain into an analog reference voltage, thereby accurately dividing the probability space and achieving quantitative control over the precision of probability adjustment. Finally, a dynamic comparison module, triggered by a clock signal, performs real-time voltage amplitude comparison between the random triangular wave signal and the control threshold signal, and outputs the initial probability bit corresponding to the desired flip probability. By capturing and latching the transient comparison results of the two signals at a specific clock edge, converting them into a single-bit signal of 0 or 1 in the digital domain, a highly statistically independent, time-aligned, and precisely configurable initial probability bit output is achieved with low hardware power consumption. Attached Figure Description

[0027] Figure 1 A schematic diagram of the structure of a PBIT generator based on random triangular waves provided in Embodiment 1 of this application;

[0028] Figure 2 A schematic diagram of the triangular wave symbol provided in this application;

[0029] Figure 3 This is a schematic diagram of the probability calculation system provided in Embodiment 2 of this application. Detailed Implementation

[0030] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0031] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0032] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0033] The following specific embodiments are given to illustrate the technical solution of this application in detail.

[0034] Figure 1 This is a schematic diagram of the PBIT generator based on random triangular waves provided in Embodiment 1 of this application. Please refer to... Figure 1 The PBIT generator provided in this embodiment includes a random signal generation module, a threshold configuration module, and a dynamic comparison module;

[0035] The random signal generation module is used to generate a random triangular wave signal with random slope characteristics based on the random data stream.

[0036] The threshold configuration module is used to provide a control threshold signal for setting the target probability resolution;

[0037] The dynamic comparison module is connected to the random signal generation module and the threshold configuration module respectively. It is used to sample and compare the random triangular wave signal with the control threshold signal under the trigger of the clock signal, and output the initial probability bits corresponding to the expected flip probability.

[0038] Specifically, a PBIT generator, also known as a Probability Bit (PBIT) generator, is a new type of hardware unit that can break through the limitations of traditional deterministic computation. Triggered by a clock cycle, it generates a binary signal that randomly toggles between logic 0 and logic 1, and the probability of its output being 1 or 0 can be precisely configured and controlled. The PBIT generator consists of three modules: a random signal generation module that generates a high-entropy analog waveform signal (i.e., a random triangular wave signal) that is irregular, unpredictable, and eliminates periodicity on the time axis; a threshold configuration module that generates one or more stable, precisely controllable static / numerically controlled analog voltage signals (i.e., control threshold signals) used as slicing standards for probability division; and a dynamic comparison module that, upon the arrival of a clock edge, collides the analog signals output from the two modules (real-time voltage comparison), discretizes and latches the continuous comparison results, and finally outputs the probability bits in the digital domain.

[0039] Furthermore, the random data stream refers to the raw, irregular binary digital signal (bit stream) generated by the hardware entropy source (such as a true random number generator, multiple random symbol pools, or a master-controlled random source) within the random signal generation module. It is the root of the randomness of the entire system. Traditional triangular waves have constant slopes on their rising and falling edges, while the random slope characteristic in this application refers to the fact that the slope, peaks, and troughs of each period, or even each symbol segment, of the triangular wave are subject to real-time abrupt changes controlled by the aforementioned random data stream. This results in a chaotic distribution of the waveform on the time axis, thus physically eliminating any residual periodicity. The random triangular wave signal refers to a continuous analog voltage waveform formed by overlapping high and low level segments with random slopes on the time axis after modulation by the aforementioned random data stream. As the random input of the dynamic comparison module, its voltage probability density distribution on the time axis is highly uniform and highly random.

[0040] It should be noted that the target probability resolution determines the fineness (step precision) of the PBIT generator's probability control. If the target probability resolution is set to 4 bits, the threshold configuration module can uniformly divide the probability from 0 to 100% into 2^4 = 16 levels (e.g., 0, 1 / 16, 2 / 16, ..., 15 / 16, 1); if the resolution is increased to 8 bits, 256 probability configuration levels can be provided. The resolution directly determines the calculation accuracy of the subsequent dynamic comparison module. The control threshold signal refers to one or more discrete DC reference voltage signals output by the threshold configuration module. The voltage amplitude of this signal has a strict mathematical mapping relationship with the desired target probability value (desired flip probability).

[0041] The initial probability bit refers to the single-bit digital signal (0 or 1) that the dynamic comparison module outputs to the subsequent digital circuit after being latched by the sampling clock (CLK). It is called initial because it is the native probability bit directly output by a single physical device, which macroscopically exhibits a specific expected flip probability. It can be directly used as the input to the subsequent circuit, or it can be fed into the cascaded operation circuit (such as AND gates or parallel counters) to evolve into a higher-order probability signal through probability multiplication or addition.

[0042] It's important to note that there is an absolute, deterministic linear (or monotonic) mapping relationship between the desired reversal probability and the control threshold signal. Since the voltage distribution of the random triangular wave in space is uniform, adjusting the voltage of the control threshold signal essentially slices this random waveform into water level segments. When the threshold voltage (water level) is set extremely low, the random triangular wave is above the threshold for most of the time, resulting in a large proportion of the dynamic comparator outputting a high-level 1, and the corresponding desired reversal probability approaches 100%. When the threshold voltage (water level) is set at the center, the random triangular wave is above and below the threshold equally, resulting in a desired reversal probability of 50%. Therefore, adjusting the amplitude of the control threshold signal allows for 100% precision in locking the desired reversal probability.

[0043] In practice, the random signal generation module performs low-level modulation on the input random data stream to generate a simulated random triangular wave signal with random slope characteristics in real time on the spatial or temporal axis, and continuously sends it to the signal input terminal of the dynamic comparison module. Simultaneously, the threshold configuration module outputs a corresponding control threshold signal (i.e., a DC reference voltage of a specific amplitude) according to the pre-set target probability resolution requirements, and stably sends it to the threshold input terminal of the dynamic comparison module. When the clock edge arrives, the dynamic comparison module acquires the two analog signals and uses its internal comparison circuit to perform real-time voltage amplitude comparison between the input random triangular wave signal and the control threshold signal. As the random triangular wave signal voltage fluctuates, the dynamic comparison module generates continuously changing transient high and low level logic based on the real-time relative height relationship between the random triangular wave signal and the control threshold signal voltage, performing high-frequency sampling and digital latching of the current transient comparison result. The latched digital signal is then output as the initial probability bit to the subsequent circuitry.

[0044] The modules will be introduced in turn.

[0045] Optionally, the random signal generation module includes: multiple independent random symbol generation sub-modules for generating multiple uncorrelated analog symbol signals in parallel; a channel selector, with multiple input terminals of the channel selector connected to the output terminals of each of the random symbol generation sub-modules; and a data selection controller connected to the control terminal of the channel selector for controlling the channel selector to dynamically and randomly switch the output of the multiple analog symbol signals according to the input control random number, thereby combining them to form a random triangular wave signal.

[0046] For details, please continue to refer to... Figure 1The random signal generation module specifically includes the following three parts: (1) Multiple independent random symbol generation sub-modules, which serve as the physical source of random waveforms. Each random symbol generation sub-module is arranged in parallel in space, operates independently, and does not interfere with each other, continuously generating multiple analog symbol signals that are statistically absolutely independent and unrelated. (2) Channel selector (usually an analog multiplexer MUX in hardware), which serves as the physical switch for waveform splicing. Its multiple input terminals are connected to the outputs of each random symbol generation sub-module, and its output terminal is connected to the subsequent dynamic comparison module. It is responsible for closing or opening specific paths under the drive of the data selection controller, thereby selecting one of the multiple analog symbol signals for output. (3) Data selection controller, which serves as the command brain for path switching. It is connected to the control terminal (gating pin) of the channel selector, receives the input random control number (original entropy source) in real time, and converts it into control commands to drive the channel selector to switch channels randomly and dynamically in different working clock cycles, thereby splicing and merging waveform segments from different sources into the final random triangular wave signal on the time axis.

[0047] Furthermore, in analog symbol signals, the analog attribute (voltage continuity) does not refer to the discrete levels of 0 or 1 in digital circuits, but rather to analog voltage signals that change continuously in both voltage amplitude and time axis. Only analog signals can subsequently be sent to the dynamic comparison module for level comparison with the control threshold signal. Moreover, a symbol represents a basic signal unit within a fixed time slice. In this application, a symbol refers to a triangular wave segment (or basic waveform unit) with a certain random slope generated by a single random symbol pool within a specific clock cycle. Analog symbol signals are essentially segments of analog triangular waves with random slopes.

[0048] In practical implementation, multiple independent random symbol generation submodules operate in physical parallel, each independently and synchronously generating one analog symbol signal and continuously supplying it to the corresponding input terminals of the channel selector. The data selection controller acquires the input control random number in real time and converts it into a gating control signal that matches the channel selector, outputting it to the control terminal of the channel selector in real time. Driven by the gating control signal, the channel selector dynamically and randomly switches its internal conduction path within different clock cycles. By performing aperiodic switching gating between multiple input analog symbol signals, the analog symbol fragments from different random symbol generation submodules are spliced ​​and fused into one signal on the time axis, and finally combined at the output terminal of the channel selector to form a random triangular wave signal.

[0049] Optionally, each of the random symbol generation submodules includes a random number generator and a symbol converter; wherein, the random number generator is used to generate a random number bit stream; the symbol converter is connected to the random number generator and is used to convert the random number bit stream into multiple sets of triangular wave symbols with different slope characteristics, as the analog symbol signal output.

[0050] Specifically, the random symbol generation submodule is responsible for completing the primary conversion from pure digital random numbers to local analog waveforms. It consists of the following two parts: (1) a random number generator, which acts as a digital generator within a single symbol pool. Driven by a high-speed clock, it is responsible for generating a pure digital domain, unpredictable, and high-entropy random number bit stream, providing the underlying randomness control source for the random symbol generation submodule. (2) a symbol converter, which is directly connected to the random number generator. It is responsible for receiving the input digital random number bit stream and, based on the transition states of 0 and 1 in the bit stream, physically mapping and converting it into continuous voltage segments (i.e., triangular wave symbols) with different rising / falling slope characteristics in the analog domain. Finally, it outputs the analog symbol signal of the random symbol generation submodule to the subsequent channel selector.

[0051] Furthermore, a random number bitstream refers to a discrete binary digital signal sequence arranged continuously on the time axis in the digital domain, with amplitudes consisting only of high-level 1 and low-level 0. The random number bitstream can be either truly random or pseudo-random. In a triangular wave symbol, a symbol refers to a basic analog signal unit within a specific and extremely small finite time slice (i.e., the characteristic frequency period). A triangular wave refers to the geometric shape in which the amplitude of its analog voltage increases linearly (positive slope) or decreases linearly (negative slope) with time within this specific time slice. In conventional circuits, the slope of a triangular wave is fixed. However, in the triangular wave symbol of this application, the slope is directly controlled by the preceding random number bitstream. For example, when the random number generator outputs the number 1, the symbol converter immediately generates a triangular wave voltage segment with a very steep slope (rapidly rising) within the current time slice; while when the output number 0, the symbol converter immediately switches to generating a triangular wave voltage segment with a gentle slope (slowly falling). These miniature segments of analog voltage, which locally exhibit a triangular shape but with varying slopes and dynamically change with the digital stream, are called triangular wave symbols. Multiple sets of such triangular wave symbols are stacked on the time axis to form multiple uncorrelated analog symbol signals.

[0052] Optionally, the triangular wave symbol undergoes a non-periodic random oscillation between a first amplitude threshold and a second amplitude threshold, and the slope feature is characterized by the transit time of the triangular wave symbol between the first amplitude threshold and the second amplitude threshold; wherein, the rising transit time of the triangular wave symbol from the first amplitude threshold to the second amplitude threshold is selectively limited to a first time threshold interval or a second time threshold interval according to the random number bit stream; the falling transit time of the triangular wave symbol from the second amplitude threshold to the first amplitude threshold is selectively limited to a third time threshold interval or a fourth time threshold interval according to the random number bit stream; the maximum value of the first time threshold interval is less than the minimum value of the second time threshold interval, and the maximum value of the third time threshold interval is less than the minimum value of the fourth time threshold interval.

[0053] Specifically, Figure 2 This is a schematic diagram of the triangular wave symbol provided in this application. Please refer to... Figure 2The triangular wave symbol is vertically confined to a non-periodic random oscillation between a first amplitude threshold and a second amplitude threshold. In this embodiment, the first amplitude threshold corresponds to the lower limit voltage of the triangular wave symbol (i.e., the trough amplitude VL), and the second amplitude threshold corresponds to the upper limit voltage of the triangular wave symbol (i.e., the apex amplitude VH), thereby safely constraining all generated analog waveform segments within a preset linear voltage working window. The slope characteristic of the triangular wave symbol is quantitatively characterized in the time domain by its transit time between the first amplitude threshold and the second amplitude threshold. Since the voltage difference between the first amplitude threshold and the second amplitude threshold is fixed, the shorter the transit time of the triangular wave symbol, the steeper the absolute value of the transient slope of the corresponding waveform; conversely, the longer the transit time, the gentler the absolute value of the transient slope of the corresponding waveform. Specifically, during the linear rise of the triangular wave symbol from the first amplitude threshold to the second amplitude threshold, the required rise transit time is controlled by the real-time state of the random number bit stream output by the preceding random number generator. When the random number bitstream is in the first random state (e.g., logic 1), the rise transit time is selectively limited to the first time threshold interval, and the waveform exhibits a high-slope, fast rising edge. When the random number bitstream is in the second random state (e.g., logic 0), the rise transit time is selectively limited to the second time threshold interval, and the waveform exhibits a low-slope, gentle rising edge. Since the maximum value of the first time threshold interval is strictly smaller than the minimum value of the second time threshold interval, the fast rising edge and the gentle rising edge do not overlap in the time domain, exhibiting significant discretization modulation characteristics. Similarly, during the linear descent of the triangular wave symbol from the second amplitude threshold to the first amplitude threshold, the required fall transit time is also modulated by the dynamic state selection of the random number bitstream. Depending on the logic state of the input random number bitstream, this fall transit time is selectively limited to the third time threshold interval (corresponding to a high-slope, fast falling edge) or the fourth time threshold interval (corresponding to a low-slope, gentle falling edge). Since the maximum value of the third time threshold interval is strictly less than the minimum value of the fourth time threshold interval, absolute isolation between the two descent rates in the time domain is ensured. Figure 2 In the triangular wave signal, symbol A corresponds to a fast rising edge and a fast falling edge with a high slope; symbol B corresponds to a fast rising edge with a high slope and a gentle falling edge with a low slope; symbol C corresponds to a gentle rising edge with a low slope and a fast falling edge with a high slope; and symbol D corresponds to a gentle rising edge with a low slope and a gentle falling edge with a low slope. This mechanism, which uses a random bit stream to randomly select the rising / falling transit time intervals without overlap, results in a highly discrete, chaotic, and unpredictable slope abrupt change characteristic in each time slice of the final combined analog random triangular wave signal. Physically, this completely eliminates the periodic residuals and spatial coupling of conventional triangular waves, greatly improving the statistical independence of the initial probability bits in subsequent cascaded operations.

[0054] Optionally, the signal generation rate of the random number generator is greater than twice the symbol transformation characteristic frequency of the symbol converter.

[0055] For details, please continue to refer to... Figure 2 The signal generation rate of the random number generator and the symbol transformation characteristic frequency of the symbol converter satisfy a specific time-series mathematical relationship, namely, the signal generation rate is strictly greater than twice the characteristic frequency. Figure 2 In the waveform time-domain distribution shown, a complete triangular wave symbol period topologically includes a complete rise transit time and a complete fall transit time, corresponding to the leading and trailing edges of the waveform, respectively. To ensure the random independence of the output waveform, it is necessary to ensure that the control signals controlling the rise and fall transits are completely decoupled within a single symbol transformation cycle. This application limits the rate of the random number generator to more than twice the characteristic frequency (T1), ensuring, from a timing matching perspective, that the number of random number bits input by the random number generator to the subsequent symbol converter is at least two within any complete characteristic cycle. This physically eliminates the undersampling of the random entropy source, ensuring that the leading and trailing edges of the waveform have completely independent, uncorrelated state selections. Simultaneously, this relationship introduces an oversampling mechanism in the time domain, allowing the symbol converter to receive multiple discrete bits within a single rise or fall transit for finer-grained secondary slope dynamic fine-tuning, greatly increasing the fragmentation and unpredictability of the triangular wave symbol on the microscopic time axis. This design breaks the inherent deterministic logic and coherence of the circuit, avoids phase locking between the waveform slope evolution and the system clock, and effectively blocks potential periodic residues. This results in the generated random triangular wave signal having extremely high and pure information entropy, providing solid underlying hardware support for the subsequent dynamic comparison module to generate highly statistically independent initial probability bits.

[0056] Furthermore, the threshold configuration module includes a digital-to-analog converter.

[0057] Optionally, the digital-to-analog converter is specifically a voltage divider resistor network used to output multiple sets of different control threshold signals, which are then output by the dynamic comparison module as corresponding multi-path constant probability bits. These multi-path constant probability bits are combined to form the initial probability bits.

[0058] Specifically, a digital-to-analog converter (DAC) is a circuit module that converts digital signals (discrete binary code) into analog signals (continuously varying voltage or current). In this application, the DAC is responsible for receiving externally input probabilistic control digital codes (e.g., an 8-bit binary number 10000000) and translating them into corresponding analog voltage thresholds (e.g., a midpoint voltage of 1.2V), which serve as the physical basis for the dynamic comparison module to slice the voltage levels. A voltage divider resistor network is a purely passive cascaded voltage divider circuit composed of multiple precisely matched resistors connected in series or parallel. When a fixed reference voltage is applied across the network, the nodes between the resistors automatically generate a series of strictly proportional, incrementally increasing / decreasing discrete analog voltage nodes. In this application, the voltage divider resistor network is used to generate multiple different control thresholds, which can simultaneously output multiple different voltage levels in hardware space without complex digital control timing.

[0059] Furthermore, multiple fixed-probability bits refer to a sequence of single-bit (1-bit) digital streams arranged in parallel in space and outputting in multiple ways. In this application, since the voltage divider resistor network simultaneously outputs multiple sets of different control threshold voltages (e.g., V1, V2, ..., Vn), the multiple parallel dynamic comparators inside the dynamic comparison module simultaneously perform multi-level collisions on the same random triangular wave. Due to the different level lines, each bit stream output after comparison has its own fixed and unique flip probability (e.g., the first bit stream has a fixed output probability of 25%, the second bit stream has a fixed output probability of 50%, and so on). Each of these independent bit streams with a fixed probability is a fixed-probability bit. Physically, the fixed-probability bits and the initial probability bits are represented as a convergence and combination relationship between a single signal line and a parallel bus / data vector: multiple fixed-probability bits are sub-units, representing the single-probability bit output of each individual path; the initial probability bits are the complete entity, a high-dimensional parallel digital signal array formed by the binding and packaging of all fixed-probability bits. For example, a voltage divider resistor network draws four different voltage lines in space (multiple thresholds). The dynamic comparison module uses four dynamic comparators to perform parallel comparisons, instantly generating four independent digital streams, each with a fixed probability characteristic (multiple fixed-probability bits). These four digital streams are combined and output to the module in a 4-bit parallel manner, as a whole, to the subsequent digital circuit. This complete 4-bit parallel probability data stream is the initial probability bit.

[0060] Furthermore, the dynamic comparison module employs a dynamic comparator.

[0061] Specifically, the dynamic comparison module employs a clock-driven dynamic comparator with a positive feedback regeneration mechanism. This naturally aligns with the clock-triggered latching requirements during the dual-stage reset and evaluation processes, exhibiting zero static power consumption, ultra-high response speed, and effective avoidance of metastability. Under this hardware architecture, multiple dynamic comparators perform real-time parallel collisions on the same high-entropy random triangular wave across multiple water level lines, thereby outputting multiple fixed-probability bits corresponding to different desired flip probabilities on each independent path. These spatially parallel single-bit digital streams are combined and packaged to ultimately form initial probability bits in the form of a high-dimensional signal array, achieving a highly efficient and low-power hardware mapping from uncertain analog waveforms to highly precise adjustable digital probabilities.

[0062] The PBIT generator provided in this embodiment, in a first aspect, generates multiple uncorrelated analog symbol signals in parallel by setting up multiple independent random symbol generation sub-modules. The data selection controller drives the channel selector to dynamically and randomly switch the output according to the control random number. The analog symbol segments from different sub-modules are spliced ​​and fused on the time axis, which can break the periodicity and spatial coupling of a single waveform. Thus, at the output of the channel selector, a high-entropy random triangular wave signal with no regularity and unpredictability in time axis and voltage amplitude is formed.

[0063] Secondly, by selectively limiting the rise and fall transit times of the triangular wave symbols to non-overlapping intervals where the maximum value is less than the minimum value according to the random number bit stream, multiple different slope characteristics are characterized. At the same time, the signal generation rate of the random number generator is limited to more than twice the symbol conversion characteristic frequency. From the perspective of timing matching, it is ensured that the number of random number bits input in any complete characteristic period is at least 2. Physically, the undersampling of the random entropy source is eliminated and the states of the waveform leading edge and trailing edge are completely decoupled and independently selected. At the same time, a time-domain oversampling mechanism is introduced to support finer-grained secondary slope dynamic fine-tuning within a single transit, thereby breaking the inherent deterministic logic and coherence of the circuit, avoiding phase locking between the waveform slope evolution and the system operating clock, effectively blocking potential periodic residues and greatly improving the statistical independence of subsequent operations.

[0064] Thirdly, the digital-to-analog converter is specifically configured as a voltage divider resistor network. Its purely passive cascaded voltage divider structure automatically generates a series of strictly proportional, incrementally increasing or decreasing discrete analog voltage nodes. Without complex digital control timing, it can simultaneously output multiple water level lines representing different control threshold signals in the hardware space. In conjunction with a dynamic comparator, it performs real-time parallel collisions of multiple water level lines on the same high-entropy random triangular wave, thereby outputting multiple fixed-probability bits with fixed and unique flip probabilities in parallel in space. These bits are then combined and packaged to form an initial probability bit in the form of a high-dimensional signal array. Finally, it efficiently realizes the hard-core hardware mapping from uncertain analog waveforms to high-precision adjustable digital probabilities with extremely low hardware power consumption.

[0065] Corresponding to the aforementioned embodiment of a PBIT generator based on random triangular waves, this application also provides an embodiment of a probability calculation system.

[0066] Figure 3 This is a schematic diagram of the probability calculation system provided in Embodiment 2 of this application. Please refer to... Figure 3 The probability calculation system provided in this embodiment includes multiple PBIT generators based on random triangular waves as described in any of the first aspects of this application, as well as probability calculation circuits.

[0067] The random sources used by each of the PBIT generators are independent of each other, and the initial probability bits output by each of the PBIT generators are independent and uncorrelated.

[0068] The probability calculation circuit is connected to the output terminal of each of the PBIT generators and is used to perform combination calculations on the initial probability bits output by each of the PBIT generators to obtain the target probability bits; the precision of the target probability bits is higher than that of the initial probability bits output by each of the PBIT generators.

[0069] Optionally, the probability operation circuit uses an AND gate as a probability multiplier and a parallel counter as a probability adder.

[0070] Specifically, the probability computation system utilizes probability data streams (bits) with specific statistical distributions output by multiple underlying hardware units (PBIT generators) as the most basic operators, directly performing complex algebraic, nonlinear function, or combinatorial optimization calculations on the hardware circuitry through logic collisions. The front-end of the probability computation system includes multiple parallel, independent PBIT generator arrays, while the back-end includes probability computation circuitry. The multiple PBIT generators are arranged in parallel arrays in space, responsible for continuously providing multiple initial probability bits to the back-end. This application particularly emphasizes that the random sources used by each PBIT generator are independent of each other. This means that although they are on the same chip, their physical entropy sources (such as the switching rules of the random symbol pool) have no inherent connection, which physically eliminates spatial coupling and crosstalk. Due to the independence of the sources, the multiple initial probability bits they output are statistically absolutely uncorrelated. This independence is the fundamental prerequisite for preventing computational distortion and mathematical failure when the back-end performs probability multiplication (AND gate) or addition operations.

[0071] Furthermore, the input of the probability operation circuit is connected to the output of all PBIT generators. It is responsible for performing hardware-level combinatorial algebraic operations (such as through AND gates, full adder trees, or parallel counters) on the received multiple independent initial probability bits, and then aggregating and integrating the results to finally output a target probability bit. Due to the limitations of the digital-to-analog converter or voltage divider resistor network, the resolution of the initial probability bit output by a single PBIT generator is limited (e.g., it can only be controlled at coarse granular levels such as 25% or 50%). The probability operation circuit, however, utilizes the principles of probability statistics to generate a high-precision target probability bit with significantly improved accuracy and resolution in the macroscopic digital domain by jointly cascading multiple independent coarse-grained initial bits.

[0072] For example, suppose the system has two completely independent PBIT generators. PBIT generator A outputs an initial probability bit with a flip probability of P_A = 0.5, and PBIT generator B outputs an initial probability bit with a flip probability of P_B ​​= 0.5. If the probability operation circuit is a simple binary AND gate, when these two independent and unrelated bit streams are simultaneously fed into the AND gate, according to the principle of probability multiplication, the flip probability of the target probability bit obtained at the AND gate output will be precisely locked as P_target = P_A × P_B = 0.25. Through this hardware-level logic cascading, the system achieves higher precision and finer-dimensional probability bit output purely through the combination of subsequent digital logic without needing to improve the manufacturing precision of the front-end analog devices.

[0073] The system in this embodiment can be used to execute... Figure 1 The steps of the device embodiment shown are similar in principle and process, and will not be repeated here.

[0074] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0075] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0076] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A PBIT generator based on random triangular waves, characterized in that, The PBIT generator includes a random signal generation module, a threshold configuration module, and a dynamic comparison module; The random signal generation module is used to generate a random triangular wave signal with random slope characteristics based on the random data stream. The threshold configuration module is used to provide a control threshold signal for setting the target probability resolution; The dynamic comparison module is connected to the random signal generation module and the threshold configuration module respectively. It is used to sample and compare the random triangular wave signal with the control threshold signal under the trigger of the clock signal, and output the initial probability bits corresponding to the expected flip probability.

2. The PBIT generator of claim 1, wherein, The random signal generation module includes: multiple independent random symbol generation sub-modules, used to generate multiple uncorrelated analog symbol signals in parallel; A channel selector, wherein multiple input terminals of the channel selector are respectively connected to the output terminals of each of the random symbol generation submodules; And a data selection controller, connected to the control terminal of the channel selector, is used to control the channel selector to dynamically and randomly switch the output of the multiple analog symbol signals according to the input control random number, and combine them to form a random triangular wave signal.

3. The PBIT generator of claim 2, wherein, Each of the random symbol generation submodules includes a random number generator and a symbol converter; The random number generator is used to generate a random number bit stream; The symbol converter, connected to the random number generator, is used to convert the random number bit stream into multiple sets of triangular wave symbols with different slope characteristics, as the analog symbol signal output.

4. The PBIT generator of claim 3, wherein, The triangular wave symbol oscillates randomly and non-periodically between a first amplitude threshold and a second amplitude threshold, and the slope feature is characterized by the transit time of the triangular wave symbol between the first amplitude threshold and the second amplitude threshold. The rise transit time of the triangular wave symbol from the first amplitude threshold to the second amplitude threshold is selectively limited to a first time threshold interval or a second time threshold interval according to the random number bit stream. The descent transit time of the triangular wave symbol from the second amplitude threshold to the first amplitude threshold is selectively limited to a third time threshold interval or a fourth time threshold interval according to the random number bit stream; The maximum value of the first time threshold interval is less than the minimum value of the second time threshold interval, and the maximum value of the third time threshold interval is less than the minimum value of the fourth time threshold interval.

5. The PBIT generator of claim 3, wherein, The signal generation rate of the random number generator is greater than twice the symbol transformation characteristic frequency of the symbol converter.

6. The PBIT generator of claim 1, wherein, The threshold configuration module includes a digital-to-analog converter.

7. The PBIT generator of claim 6, wherein, The digital-to-analog converter is specifically a voltage divider resistor network, used to output multiple sets of different control threshold signals. The dynamic comparison module outputs corresponding multi-path fixed probability bits, which are combined to form the initial probability bits.

8. The PBIT generator of claim 1, wherein, The dynamic comparison module employs a dynamic comparator.

9. A probabilistic operation system, characterized by, The probability calculation system includes a plurality of PBIT generators based on random triangular waves as described in any one of claims 1 to 8 and a probability calculation circuit; The random sources used by each of the PBIT generators are independent of each other, and the initial probability bits output by each of the PBIT generators are independent and uncorrelated. The probability calculation circuit is connected to the output terminal of each of the PBIT generators and is used to perform combination calculations on the initial probability bits output by each of the PBIT generators to obtain the target probability bits; the precision of the target probability bits is higher than that of the initial probability bits output by each of the PBIT generators.

10. The probabilistic operation system of claim 9, wherein, The probability operation circuit uses AND gates as probability multipliers and parallel counters as probability adders.