Touch Data Processing with ML Accelerator for Power Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing touch-enabled computing devices face challenges in efficiently determining intentional touch inputs, leading to unnecessary processor wake-ups and high power consumption due to accidental touches, such as palm contacts or object activation.
Innovation Solution
The implementation of a low-power machine learning (ML) accelerator within the embedded subsystem circuitry, which processes touch data from touch sensors to differentiate between intentional and unintentional inputs, reducing the need for processor intervention and minimizing power usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the processor continuously monitors touch data to determine intentional inputs, then touch input accuracy is improved, but power consumption increases due to unnecessary processor wake-ups
Solution Approach 1:
The patent introduces an embedded subsystem as an intermediary between the touch sensor and the main processor. This subsystem pre-processes touch data and determines whether touches are intentional before waking the processor, thereby maintaining accurate touch input detection while reducing processor power consumption by keeping it in sleep mode during low-power states.
Solution Approach 2:
The patent divides the touch processing system into two separate components: an embedded subsystem for initial touch data processing and intentionality determination, and a main processor for executing applications. This segmentation allows the embedded subsystem to handle power-consuming processing tasks while the main processor remains in low-power state, resolving the contradiction between continuous monitoring accuracy and power consumption.
2Speed
If the processor wakes up frequently to process all touch data, then touch responsiveness is improved, but power consumption increases due to unnecessary wake-ups from accidental touches
Solution Approach 1:
The embedded subsystem performs preliminary processing of touch data before the main processor needs to wake up. By determining touch intentionality in advance at the embedded subsystem level, the system avoids unnecessary processor wake-ups while ensuring that genuine touches are responded to quickly, thus maintaining responsiveness while reducing power consumption.
Solution Approach 2:
The embedded subsystem acts as a mediator that filters touch data before it reaches the main processor. It quickly evaluates whether a touch is intentional and only wakes the processor when necessary, thereby maintaining fast response times for valid touches while eliminating power-wasting wake-ups caused by accidental touches.
3Productivity
If a dedicated microcontroller is used to process touch frames, then processing capability is improved, but device complexity and cost increase
Solution Approach 1:
The patent utilizes the application processor's embedded subsystem, which is already present for other purposes (such as audio processing), to also handle touch data processing. This multi-functional use of the embedded subsystem provides dedicated touch processing capability without adding separate dedicated microcontroller hardware, thereby maintaining high processing capability while avoiding increased device complexity and cost.
Data Source
AI summary
Methods, apparatus, systems, and articles of manufacture are disclosed to process touch data. An example apparatus includes machine learning accelerator circuitry to execute a machine learning algorithm on touch data from touch sensor circuitry; and determine, based on an output of the machine learning algorithm, whether a touch input corresponding to the touch data was intentional; transceiver circuitry to, after a determination that the touch input was intentional, provide touch coordinates to memory; and processor circuitry to, after the determination that the touch input was intentional: access the touch coordinates in the memory; and perform an action based on the touch coordinates.