TSMC N12e: Powering the Next Generation of AI and 5G era IOT Edge Devices

With tens of billions of Internet of Things (IOT) devices already deployed throughout the world, it is hard to describe IOT as an emerging category. International Data Corporation (IDC) estimates that by 2025, there will be over 40 billion connected IOT devices, generating nearly 80 zettabytes of data [1]. Clearly, the number of devices does not suggest a nascent market but the 5G and AI megatrends will transform IOT devices and push device capabilities to realize true edge computing.

The first wave of IOT devices in the market today gained mass adoption because they had offered new functionalities not previously available – connected speakers, smart doorbells, connected security cameras, smartwatches, wearables etc. These devices are innovative and are pushing technology boundaries but had some technology limitations. For IOT devices, key design considerations include form factor, power efficiency, battery life, performance and connectivity.

TSMC's N12e brings the power of FinFET to 5G and AI enabled IoT Edge Devices TSMC's N12e brings the power of FinFET to 5G and AI enabled IoT Edge Devices

Smart earbuds sound great and offer new functionality but cannot last the duration of an overseas flight. Connected security cameras send gigabytes of HD video to the cloud on a false trigger (look, squirrel!) of an alert event. Humans are still modifying their voice and speech syntax so that their smart speakers can understand them. People with smartwatches have battery anxiety if they forget their charging cables on an overnight trip.

The next generation of AI and 5G IOT devices will be transformative and deliver new levels of intelligence, functionality, connectivity, reliability and performance at the edge. Ever more powerful and sophisticated AI neural networks are being trained to better mimic the human brain. These deep neural networks (DNNs) and convolutional neural networks (CNNs) will give edge devices capabilities beyond the IOT devices of today.

Natural Speech Recognition – people can speak naturally and have their device understand them. By moving the AI inferencing to the edge, the response latency will improve greatly as well as enhanced functionality when not connected to the cloud

Enhanced Machine Vision – insects, shadows or animals often falsely trigger connected security cameras. By moving the image classifier to the edge, the AI-enabled connected cameras can continually monitor for humans – even with facial recognition but ignore pets and insects without sending gigabytes of HD video into the cloud for inferencing

Smart Health Monitoring – AI has become much more pervasive in healthcare, often providing better diagnosis accuracy versus human medical professionals. As smartwatches enhance their sensor packages and performance they can use AI inferencing at the edge for better monitoring

5G is another megatrend that will affect many technology areas. In the context of IOT, 5G will bring enhanced connectivity.

Device Connection Density – 5G networks can support up to 1 million devices per 0.38 square miles versus 2000 for 4G networks [2].

Device Latency – 5G networks will offer significantly better latency over 4G networks, theoretically as low as 1 millisecond, an order of magnitude improvement over 4G [3]. Latency is particularly important for time sensitive workloads, like enhanced car safety or industrial automation.

AI and 5G enhanced IOT devices will be diverse by type and functionality. The common requirements for this new generation of devices are the demand for more compute performance, and equally important, improved power efficiency and longer battery life. The semiconductor technology must advance – this is foundational for 5G and AI enhanced IOT devices.

Planar transistors, generally 20/22nm technology or above, have been the foundation for all semiconductors for the past 50 years. Planar transistors have long reached the limits of scaling, because much better control of the off-state leakage current by the gate is required for short gate length transistors. FinFET transistors are a quantum leap forward for scaling, performance, power efficiency and leakage. The most advanced TSMC node in production today, the N5 family, is FinFET technology.

N12e's FinFET faster performance and power efficiency including better leakage current N12e's FinFET faster performance and power efficiency including better leakage current

TSMC developed N12e specifically for AI-enabled IOT and other high efficiency, high performance edge devices. N12e brings TSMC’s world class FinFET transistor technology to IOT. N12e is a significantly enhanced technology derived from the lineage of TSMC’s 16nm FinFET technology first introduced in 2013. Through years of process development and enhancements, N12e is based on the TSMC 12FFC+_ULL technology. The TSMC 16/12 nm family of technologies are in today’s supercomputers and high-performance computing devices like GPUs and Network Processors. TSMC has brought this brawny performance to IOT with N12e.

Compared to TSMC’s 22ULL, N12e offers:

  • 76% improvement in logic density – allowing smaller and more cost effective designs or, in the same given area, pack many more transistors for added compute cores and memory
  • 49% improvement in speed at a given power – a big leap forward for IOT devices at any given power level. N12e has much higher headroom for frequency and drive power to deliver much more performance over planar technologies
  • 55% improvement in power consumption at a given speed – N12e delivers a broad range of performance-to-power options allowing for use in many different product designs
  • More than 50% reduction in SRAM leakage current – critical for improved battery life as well as reduced heat generation and thermal dissipation
  • Low Vdd Design Ecosystem Solution – Reducing both active power and leakage power for battery-operated products. N12e can support 0.4V operation

Adapting TSMC’s 16/12 nm ultra-high performance FinFET technology for IOT products delivered the obvious performance benefits but required us to tune our recipe to deliver improved power efficiency and lower leakage. Special attention was paid to the off-state leakage characteristics.

As IOT devices evolve for the 5G and AI era, they require new fundamental semiconductor technology to support the need for faster performance, improved power efficiency and improved power leakage. While these requirements may seem at odds with each other, TSMC has been able to deliver the optimal balance with N12e. We are excited for the new class of IOT devices that 5G and AI will enable.

Please learn more at n12e.tsmc.com

All diagrams, animations and videos are for demonstrative and illustrative purposes only

References
  1. International Data Corporate (IDC), "The Growth in Connected IoT Devices Is Expected to Generate 79.4 ZB of Data in 2025," 19 June 2019. [Online]. Available here.
  2. L. Notwell, "CIO.COM," CIO.COM, 2 November 2017. [Online]. Available here.
  3. S. Hill, "5G vs. LTE: What's the difference, and does it matter?," Digital Trends, 5 December 2019. [Online]. Available here.

Related Information