TSMC new N4X process plans to max out HPC performance in 2023
TSMC’s N4X node is the company’s first High Performance Computing Focused Process Technology
TSMC’s largest growth vector is the HPC market, with companies like AMD, Nvidia, Apple, and Intel all being major customers of their latest lithography technologies. With this in mind, TSMC has revealed their first “HPC-focused” process technology, N4X, which is due to start risk production in the first half of 2023.
In a sense, N4X is an enhancement of TSMC’s N4P technology, which itself is a more advanced version of TSMC’s N5 (5 nm) node. When compared to N5, N4X should enable performance boosts of 15%, and when compared to N4P, N4X should enable a 4% boost for users. More importantly, N4X will allow TSNC’s customers to aim for higher clock speeds and utilise higher voltages more effectively, potentially enabling a larger performance boost than TSMC’s already listed 15%.
With N4X using “common design rules” with N5, TSMC’s customers should be able to transition many of their products to TSMC’s HPC-focused node without many challenges, creating an easy avenue towards increased HPC performance. This should allow refreshed TSMC N5 products to be created using N4X, allowing TSMC’s customers to create faster products with relatively little effort. TSMC has also optimised their N4X technology with their 3DFabric techniques in mind, making N4X an attractive technology for customers who plan to utilise advanced packaging technologies.
Below we have included TSMC’s N4X press release and information from a TSMC blogpost that also discusses their HPC-focused process technologies.Â
PR – TSMC Introduces N4X Process
TSMC (TWSE: 2330, NYSE: TSM) today introduced its N4X process technology, tailored for the demanding workloads of high performance computing (HPC) products. N4X is the first of TSMCâs HPC-focused technology offerings, representing ultimate performance and maximum clock frequencies in the 5-nanometer family. The âXâ designation is reserved for TSMC technologies that are developed specifically for HPC products.
Leveraging its experience in 5nm volume production, TSMC further enhanced its technology with features ideal for high performance computing products to create N4X. These features include:
– Device design and structures optimized for high drive current and maximum frequency
– Back-end metal stack optimization for high-performance designs
– Super high density metal-insulator-metal capacitors for robust power delivery under extreme performance loads
These HPC features will enable N4X to offer a performance boost of up to 15% over N5, or up to 4% over the even faster N4P at 1.2 volt. N4X can achieve drive voltages beyond 1.2 volt and deliver additional performance. Customers can also draw on the common design rules of the N5 process to accelerate the development of their N4X products. TSMC expects N4X to enter risk production by the first half of 2023.
âHPC is now TSMCâs fastest-growing business segment and we are proud to introduce N4X, the first in the âXâ lineage of our extreme performance semiconductor technologies,â said Dr. Kevin Zhang, senior vice president of Business Development at TSMC. âThe demands of the HPC segment are unrelenting, and TSMC has not only tailored our âXâ semiconductor technologies to unleash ultimate performance but has also combined it with our 3DFabricâ¢ advanced packaging technologies to offer the best HPC platform.â
TSMCâs HPC platform not only offers performance-optimized silicon with N4X technology, but also provides the greatest design flexibility with its comprehensive 3DFabricâ¢ advanced packaging technologies and a broad design enablement platform with our ecosystem partners through the TSMC Open Innovation PlatformÂ®.
TSMC Blog – The Future is Now
Evolving Computing Landscape
When I worked for a big corporate executive in charge of mainframe business in the early 2000s, we spent a lot of time visiting Fortune100 CIOs in New York City. And while mainframes still are strategically critical today, I look back and realize how rapidly the computing landscape has shifted, and how the future is now.
Distributed systems took over many computing workloads over the last few decades. Then the mobile revolution put computing into billions of devices in our pockets. Today, cloud computing and accelerated computing bring compute power to billions of people and trillions of intelligent machines. Ultimately, the boundary between the physical and virtual world will likely blur over the next decade!
And the underpinning of all of this? Semiconductor technology.
On a late night in a university lab, one of my classmates asked me a question: âIf the computer I have now is good enough for all my needs, why do we need to push for device scaling and new chips?â We clearly lacked imagination on how many new applications could be enabled as computing power skyrocketed over the coming years.
Since the emergence of the Internet and the smart phone, thereâs been an explosion of data. In 2020, over 60 zettabytes (ZB) were created or replicated, according to IDC (IDC Global DataSphere and StorageSphere Forecasts, 2021). An insatiable amount of high-performance computing is required to filter and process all of it.
High performance technology, driven by leading edge process nodes, is the engine for the next level of innovation. Take machine learning as an example. It isnât new. The fundamental concept actually dates back to the 1940s. Largely thanks to significant leaps in computer performance, previously unimaginable progress has been made in recent years. Computing power, measured by floating-point operations per second, or FLOPS, has improved by five orders of magnitude in the last 20 years! And that pace will continue to accelerate.
N4X Technology Tailored for Performance
For years, TSMC has steadily grown our emphasis on performance, providing the semiconductor technology to help power our HPC customersâ innovation. Weâve put all compute-intensive applications – server CPUs, client CPUs, GPUs, AI, and network processors – on our High-Performance Computing Platform.
HPC products have the following unique attributes:
– Higher performance and usually higher frequencies
– Power consumption on the order of 100watts â approaching 1000watts in extreme cases
– Heavier utilization and a higher percentage of dynamic power in the power envelope
– More SRAMs (>1Gb) on SoC
– Higher memory bandwidth
– Higher speed IO connectivity
– Larger die sizes with challenges to manufacturability and yield
Today, we introduced our N4X process technology, tailored for these extreme demands of high-performance computing. N4X will be the first of TSMCâs HPC-focused brand, with the âXâ designation representing ultimate performance and maximum clock frequencies. Risk production for N4X is expected by the first half of 2023.
What sets N4X apart is the optimization of the FinFET transistor and back-end-of-line process for overdrive conditions. Here is an overview of what N4X offers:
– Up to 15% performance boost over N5 (at supply voltage of 1.2V) â transistor performance as a function of voltage is optimized, with a slight tradeoff in leakage current
– Higher overdrive voltages beyond 1.2V to unlock additional performance
– Lower resistance and parasitic capacitance of targeted metal layers – back-end-of-line metal layer optimization greatly affects HPC products, due to larger die sizes, higher clock frequencies, and higher operating voltages
– Super-high-density metal-insulator-metal capacitor for most effective and reliable power delivery – Depending on product design, this performance element can minimize supply voltage droop under high current loading and increase product performance by 2~3%.
Power of the Platform
With increasing computing demand, we are seeing HPC chip designs approaching the maximum reticle size even with most leading-edge semiconductor process nodes. Fortunately, TSMC not only offers performance-optimized silicon with N4X technology, but we also provide a comprehensive HPC platform.
TSMC Open Innovation PlatformÂ® (OIP), a comprehensive design enablement platform, offers foundational IPs as well as a wide variety of high-performance IPs with our ecosystem partners.
Using TSMCâs 3DFabricâ¢, our customers can expand the number of leading-edge compute chips for maximum compute power. Or, chips can be partitioned into multiple chiplets with each adopting the optimum technology of choice â logic optimized, IO and analog optimized, or memory optimized. This adds additional dimensions to the overall system performance optimization. Furthermore, System on Integrated Chips (TSMC-SoICÂ®) provides our customers capability to achieve monolithic-like interconnect density, when compared to micro-bump based designs.
Within the same 3DFabricâ¢ package, specialty chiplets can be integrated along with leading edge high performance chips. For example, additional deep trench capacitors, or high-speed memories, and even optical IO chiplets can be integrated at close proximity with high routing density. To effectively address memory bandwidth, power delivery, or high-speed IO, the possibility of innovation for 10X like performance improvement is endless.
This is the power of the TSMC HPC platform!
Looking into the Future
Computing power has transformed society, impacted every industry, and touched every part of our lives. We are witnessing something truly remarkable and unprecedented in this golden era of high-performance compute. N4X is only the first of TSMCâs extreme performance enhancements, as we are ready to support our customersâ innovations for many generations to come.
Our HPC platform is built on TSMCâs long-term research and development in advanced technology, and our close engagements with many HPC customers in recent years. We will continue to strengthen our platform offerings that include process and advanced packaging technologies, state-of-the-art manufacturing, and design enablement solutions, all targeted to helping our customers capture the growing HPC opportunities. This is not just a promise for the future. With the introduction of the N4X extreme performance process technology, the future is, indeed, now.