Photoelectric In-Memory Computing Array for Optical Signal Multiplication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computers with Von Neumann architecture face energy inefficiencies and slow operation speeds due to separate memory and computing units, particularly in processing complex algorithms like neural networks, and existing in-memory computing devices like RRAMs and FLASHs have yield and storage limitations.
Innovation Solution
A photoelectric computing device utilizing semiconductor materials to perform in-memory computing, storing optical signals for long periods and enabling multiplication with a single device, suitable for accelerating algorithms like neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If Von Neumann architecture with separate memory and computing units is used, then device structure is simple and manufacturing is easy, but energy consumption is high and operation speed is slow due to repeated data transmission
Solution Approach 1:
The patent merges the memory unit and computing unit into a single integrated device. The semiconductor device stores optical signals in a depletion region and performs multiplication operations in-situ, eliminating the need for separate memory and computing units. This integration resolves the contradiction by combining storage and computation functions within the same physical structure, reducing energy consumption from data transmission while maintaining manufacturability through standard semiconductor processes.
Solution Approach 2:
The patent introduces an optical signal as an intermediary carrier between storage and computation. Optical signals are used to encode data that is stored in the depletion region and then read out to perform multiplication operations. This optical intermediary enables efficient data transfer and processing without the energy overhead of traditional electrical signal transmission between separate memory and computing units.
2Measurement precision
If conventional multipliers with tens of thousands of transistors are used, then multiplication precision is high, but device complexity is high and energy efficiency is low
Solution Approach 1:
The patent replaces the mechanical/electrical transistor-based multiplication system with an optical-based system. Instead of using tens of thousands of transistors to perform multiplication, the device uses optical signals to encode data and a depletion region to store and process these signals. The multiplication operation is achieved through optical readout and electronic signal processing, dramatically reducing device complexity from O(10^4) transistors to a single semiconductor device while maintaining precision.
3Quantity of substance
If RRAMs are used as in-memory computing devices, then storage capability is improved, but manufacturing yield and uniformity deteriorate due to inability to use standard CMOS process
Solution Approach 1:
The patent changes the operational parameters and material requirements to be compatible with standard CMOS processes. Instead of using RRAM materials and structures that require specialized manufacturing, the device uses a standard semiconductor substrate with a depletion region created through conventional doping and biasing techniques. The storage capability is achieved by controlling the electrical and optical parameters of the depletion region rather than relying on non-standard RRAM materials, thus maintaining high manufacturing yield and uniformity.
4Quantity of substance
If FLASHs are used as in-memory computing devices, then storage density is improved, but storage precision deteriorates due to difficulty in achieving multi-level storage
Solution Approach 1:
The patent introduces dynamic control of the depletion region to achieve precise multi-level storage. By dynamically adjusting the bias voltage applied to the semiconductor substrate, the depth and properties of the depletion region can be controlled to store optical signals with different intensity levels. This dynamic control enables precise storage of multiple bits per device, overcoming the static limitation of conventional FLASH memory while maintaining high storage density.
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
The device achieves high-precision in-memory computing, storing optical signals for extended durations and performing multiplication efficiently, reducing energy consumption and enhancing operation speed.
Implementation Method 1
applying a voltage to the substrate or applying a voltage to the control gate, so that a depletion layer is formed in the substrate... configured to store the optical signals for a long period of time
Data Source
Figure 1~2
Figure 3~4
Figure 5~7
AI summary
A photoelectric computing unit, a photoelectric computing array and a photoelectric computing method. The photoelectric computing unit comprise a semiconductor multifunctional region structure, which comprises at least one carrier control region, at least one coupling region, and at least one photon-generated carrier collection region and readout region.