MEMS Microphones: The Invisible Infrastructure of the AI Era


The era of "touch," where the world was controlled by the fingertips, is fading, and the era of "voice," where everything is commanded with the voice, is rapidly dawning. The generative AI revolution, sparked by ChatGPT, is now extending beyond PCs and smartphones to permeate all devices, evolving into personalized AI agents. Amid this massive shift, a key bottleneck that will determine the success of the on-device AI revolution has emerged: how well machines listen to human speech. This is more than a simple technical challenge; it is a fundamental prerequisite that will determine the success or failure of the AI era.

The key to this massive technological shift lies, surprisingly, in a component smaller than a fingernail: the MEMS (Micro-Electro-Mechanical Systems) microphone. And the answer lies in Infineon, the German semiconductor company that holds a significant market share in this market as of 2023. In the age of AI, where does the future of "listening technology" lie? We interviewed Jaewook Shin, Infineon's Asia Pacific Marketing Manager, to find out.

1. AI lost in the noise, 'voice' was the problem.

Why is microphone technology receiving renewed attention in the age of artificial intelligence? The reason is clear. This is because the performance of an AI system is entirely dependent on the quality of the input data. Especially in the case of voice AI, inaccurate and noisy voices can be a fatal weakness that can paralyze the entire system. Even the most advanced AI model will inevitably produce incorrect answers if it fails to properly listen. The age-old computer science adage, "Garbage in, garbage out," has become even more critical in the AI era.

Existing microphone technology has shown clear limitations in capturing faint voices from a distance or accurately isolating user commands in complex noisy environments like noisy cafes. Inaccurate voice input goes beyond simple recognition errors, leading to a chain of problems that increase the computational burden on AI systems and reduce power efficiency.

Regarding these market changes, Shin Jae-wook, the person in charge, diagnosed them as follows:

"With the recent advancements in AI solutions becoming so seamless, the need to open an app and log in to search for something will become less and less frequent. Ultimately, devices that accurately understand voices will become increasingly important."

Ultimately, fully harnessing the intelligence of AI requires a breakthrough in the performance of microphones, the most fundamental stage of sound processing: converting voice into data. Let's delve into the core of how Infineon is addressing this challenge with its unique technology.

2. Infineon's Answer: Two Technologies Capturing the Essence of Sound

Infineon's unrivaled leadership in the MEMS microphone market, which it holds by taking a significant portion of, is due to its unique technological IP (intellectual property) that goes beyond simple economies of scale. Infineon offers two core MEMS microphone structures to address the diverse needs of the market.

SBP (Single Backplate): While this is a widely used standard technology, Infineon has added its own unique "Flap Structure" and "Landing Pads" design to maximize performance and reliability. This solution is optimized for a variety of applications where robustness and stability are critical.

SDM (Sealed Dual Membrane): This next-generation technology, based on Infineon's proprietary IP, is unique. This unique structure, which literally utilizes two membranes, effectively blocks external noise while capturing the human voice with the utmost purity and clarity. This is Infineon's 'secret weapon' that achieves the best acoustic performance, that is, high signal-to-noise ratio (High SNR).

Blocks out external noise and provides highly reliable voice input.

Mr. Shin Jae-wook emphasized the importance of SDM technology in particular.
"SDM technology is a unique competitive advantage that only Infineon possesses compared to its competitors, and it is a core technology that leads the market." Beyond a simple technical concept, concrete data demonstrates how dramatically this unique technology makes a difference in actual AI performance.

3. 'Further, Clearer' - The Power of High SNR Proven in Numbers

Technological superiority ultimately only becomes valuable when proven in real-world applications. Infineon clearly demonstrates the impact of its high-SNR microphone technology on AI speech recognition through an objective metric called the "Word Error Rate (WER)." WER measures how accurately AI converts human speech into text, with lower numbers indicating better performance.

Infineon's internal tests clearly demonstrate the difference microphone performance makes to real-world recognition rates. The table below simulates a situation where the sound pressure level (dBA) reaching the microphone decreases as the distance between the speaker and microphone increases.