Parallel Analog In-Memory Computing with Current Replication Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computing systems face limitations in scalability, precision, and robustness due to energy losses, bandwidth constraints, and environmental factors, particularly in analog and current-domain computing systems, which suffer from line resistance increases and temperature-induced drift, leading to precision degradation and power consumption issues.
Innovation Solution
A method and device for parallel current-domain analog in-memory computing that replicates and processes analog current signals using current replication and switch modules, performing weighted operations according to Kirchhoff's current law, eliminating the need for digital signal conversion and enabling efficient parallel processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional current-domain computing is used, then analog computing capability is achieved, but line resistance increases causing precision degradation
Solution Approach 1:
The patent transforms the computing paradigm from current-domain to voltage-domain by introducing dual-rail voltage signals. This dimensional change allows the system to overcome line resistance effects because voltage signals are less susceptible to resistive voltage division errors compared to current signals. The dual-rail architecture encodes information in voltage differential pairs, enabling precise computation despite increasing line resistance in large-scale arrays.
Solution Approach 2:
The patent changes the fundamental signal parameter from current to voltage. By using voltage-domain computing with dual-rail signaling, the system alters the physical parameter used for information representation. This parameter change directly addresses the line resistance issue because voltage signals can be regenerated and are not subject to the same ohmic losses as current signals, thereby maintaining computational precision in large-scale implementations.
2Quantity of substance
If memory array scale is expanded, then computing capacity increases, but temperature drift causes precision loss
Solution Approach 1:
The patent employs differential signaling where positive and negative rails act as counterweights to each other. The dual-rail voltage representation uses complementary signals where one rail increases while the other decreases, creating a balanced system that rejects common-mode temperature drift. This anti-weight approach allows the system to maintain precision despite temperature variations affecting individual memory elements, as the differential measurement cancels out systematic drift.
Solution Approach 2:
The patent uses asymmetric voltage encoding where information is represented by the differential voltage between two rails rather than absolute voltage levels. This asymmetric representation in the voltage domain provides immunity to temperature drift because the relative voltage difference remains stable even when absolute levels shift due to thermal effects. The asymmetric encoding scheme enables robust computation across varying temperatures.
3Power
If Von Neumann architecture is used, then processing capability is achieved, but data transmission causes energy loss
Solution Approach 1:
The patent merges the storage and computation functions into a unified in-memory computing architecture. By performing voltage-domain computations directly within the memory array using crossbar structures, the system eliminates the need to transfer data between separate memory and processing units. This merging of functions removes the energy-consuming data transmission step while maintaining full processing capability through in-situ matrix operations.
Solution Approach 2:
The patent introduces voltage-domain signaling as an intermediary that enables direct computation within the memory array. The dual-rail voltage signals serve as mediators that can be processed directly in the memory crossbar without conversion to current domain or digital domain. This intermediary voltage representation allows energy-efficient in-memory computing by eliminating multiple signal conversion steps and long-distance data transfers required by Von Neumann architecture.
4Measurement precision
If digital conversion is performed, then signal processing accuracy is improved, but conversion time and power consumption increase
Solution Approach 1:
The patent maintains continuous analog voltage-domain processing throughout the computation pipeline without interruption for digital conversion. The dual-rail voltage signals can be processed, transformed, and computed upon continuously in the analog domain, eliminating the discrete conversion steps that interrupt the computational flow. This continuity preserves signal integrity and accuracy while eliminating the time penalties associated with analog-to-digital and digital-to-analog conversions.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Achieves ultra-large-scale integration with precise computational results, reduces line resistance impacts, and maintains high precision and robustness across varying temperatures, enhancing computational efficiency and reducing power consumption.
Implementation Method 1
replicating the analog current signal to form a corresponding replicated current signal
Implementation Method 2
matrix-vector multiplication calculations are performed in situ at output nodes according to Kirchhoff's law
Implementation Method 3
performing weighted accumulated operation of the set of modulated current signals according to Kirchhoff's current law to obtain an output current signal
Data Source
AI summary
A method for parallel analog in-memory computing is provided. The method includes the following steps: inputting an analog current signal; replicating the analog current signal to form a corresponding replicated current signal, and performing weighted processing of all the replicated current signals to obtain a corresponding set of modulated current signals; and performing weighted accumulated operation of the set of modulated current signals according to Kirchhoff's current law to obtain an output current signal. In the present disclosure, input, processing and output of signals are performed in a pure current domain, and precise replication and output of current signals are achieved, thereby ensuring the output current precision even under the condition of a high line resistance.