PWM Stochastic Computing Circuits Without On-Chip Random Generators
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Stochastic computing systems face high energy consumption and hardware costs due to the need for stochastic bit streams generated by physical random sources or pseudo-random number generators, which contribute significantly to the overall cost and power consumption.
Innovation Solution
The use of pulse-width modulated (PWM) signals as analog periodic pulse signals to encode data values, allowing for energy-efficient stochastic processing by adjusting frequency and duty cycles, enabling operations with conventional stochastic digital logic components without the need for random or pseudo-random digital bit streams.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If stochastic bit streams are generated using physical random sources or pseudo-random number generators, then stochastic computing operations can be performed, but hardware cost and power consumption increase significantly
Solution Approach 1:
The patent extracts and removes the stochastic number generator (SNG) modules from the stochastic computing system. By using externally provided pseudo-random bit streams instead of on-chip SNGs, the design eliminates the hardware components that consume significant power, thereby resolving the contradiction between maintaining stochastic computing capability and reducing power consumption
Solution Approach 2:
The patent makes the stochastic processing unit universally applicable by accepting pseudo-random bit streams from external sources rather than requiring dedicated on-chip random number generators. This multi-functionality approach allows the same processing unit to operate with bit streams generated by various external methods, reducing hardware overhead and power consumption while maintaining computational capability
2Reliability
If stochastic number generator modules are included in the system, then stochastic bit streams can be generated, but hardware cost increases heavily
Solution Approach 1:
The patent extracts and removes the stochastic number generator (SNG) modules from the stochastic computing system. By using externally provided pseudo-random bit streams instead of on-chip SNGs, the design eliminates the hardware components that consume significant power, thereby resolving the contradiction between maintaining stochastic computing capability and reducing power consumption
Solution Approach 2:
The patent introduces an intermediary interface that accepts pseudo-random bit streams from external sources. This intermediary approach allows the system to obtain necessary stochastic bit streams without requiring complex on-chip random number generation hardware, thereby reducing hardware cost while maintaining the availability of stochastic bit streams for processing
3Quantity of substance
If conventional binary radix representation is used, then compact data representation is achieved, but error tolerance decreases due to high-order bit weighting
Solution Approach 1:
The patent changes the fundamental parameter of data representation from conventional binary radix to stochastic representation. In stochastic representation, data values are encoded as probabilities in random bit streams rather than fixed-position binary digits. This parameter change equalizes the weight of all bits, making the system tolerant to bit flips while maintaining computational functionality through probabilistic encoding and processing
Data Source
AI summary
Devices and techniques are described in which stochastic computation is performed on analog periodic pulse signals instead of random, stochastic digital bit streams. Exploiting pulse width modulation (PWM), time-encoded signals corresponding to specific values are generated by adjusting the frequency (period) and duty cycles of PWM signals. With this approach, the latency, area, and energy consumption are all greatly reduced, as compared to prior stochastic approaches. Circuits synthesized with the proposed approach can work as fast and energy efficiently as a conventional binary design while retaining the fault-tolerance and low-cost advantages of conventional stochastic designs.