Before the 90's, TSMC started manufacturing semiconductors with a 3 micron process . Little more than 20 years later, the Taiwanese manufacturer brings to the world its most successful nodes: the 7nm and 5nm. We review its history to understand how this important aspect of chips has evolved.
We consider the history of the evolution of the process of processors in computing to be interesting. We are always talking about lithographs, dies, nanometers, etc., in the news, so we can bring a deeper insight into it. TSMC is the king in semiconductor manufacturing, so let's see the evolution of its process from the beginning.
1987, the founding of TSMC and the start with 3 microns
From the outset, the Taiwanese manufacturer wanted to implement an exclusive internal R&D strategy, which would be key in today's times. The company was founded in 1987 and began manufacturing chips from 2µm (microns) to 3.5µm. In pursuit of the cutting edge for more advanced processes, TSMC collaborated closely with the Taiwan Industrial Technology Research Institute.
Spend 1 year and TSMC gets to launch a process of 1.5μm, which continued during the nineties until the first process of 0.18μm of the world, which was presented as "low consumption". All of this was possible thanks to its internal R&D strategy.
1999, 0.18µm: the world's first energy-efficient technology
The 90s were abuzz with innovations and technological advancements, so manufacturers like TSMC had to accelerate their newest technology. In this way, the 0.18 micron process arrived in 1998, but in just 2 years they would go down to nanometers.
1 micron = 1000 nanometers
The Taiwanese made use of FinFET technology, which they have used to this day. Low consumption chips were in demand to be able to give life to smart devices, and especially the telephony sector that was beginning to expand.
2001, 0.13µm arrives in SoCs
Following demands from TSMC partners about processes tailored for SoCs, the manufacturer launched the first 0.13 micron process - Low Power for SoCs. At that time, we saw Nintendo and Sony consoles that came with these chips called System on a Chip.
In this year, TSMC decided to relocate its R&D team to the Hsinchu offices for the successful development of cutting-edge technology. As a result, Taiwan was prepared to host sufficient entrepreneurial capacity to develop high technology.
TSMC made use of CMOS transistors and tiny SRAM memory to make these 0.13 micron SoCs. We would soon see the first chips with a 193nm lithography, which would be the first 8-layer low-k chips and copper wire.
This advance meant an increase in power in computers, mobiles, the automotive sector, IoT or portable devices (PDAs, laptops, telephones, etc.).
2004, TSMC is the first to launch the 90nm process
Indeed, the Taiwanese manufacturer became the first to introduce the 90nm process using immersion lithography technology. The announcement would be made in Japan, at the SEMICON event that was held in December 2004.
We came from the 193nm and 157nm, which already brought news in production: the former were manufactured using a water media scanner, but the latter were produced with a dry scanner. The most remarkable thing about this is that TSMC broke the barriers defined by Moore's Law.
All of this TSMC-generated movement in the tech industry was observed by companies like GlobalFoundries (not founded until 2009), but Samsung would take longer to get into business with Samsung Foundry: 2017.
2005, the evolution of TSMC was unstoppable, 65nm!
Only 1 year later, TSMC begins to produce chip samples with a 65nm process, although we would have to wait for the approval of the product in 2006. The news that 65nm brought was interesting:
- It was a 3rd generation process.
- It integrated low-power dielectric and copper interconnections.
- Cell gate density is twice as large as 90nm.
- Higher performance and integration.
- The appearance of the 65nm LP (Low Power) process for more efficient chips.
- It served many, many uses: eDRAM, flash memory (eFLASH), high voltage, as well as processors, SoCs, etc.
Additionally, the 55nm LP and ULP (Ultra Low Power) process was introduced, providing an improved PPA along with a retractable DIE size.
2008, 40nm mass production
The Taiwanese manufacturer was once again the leading-edge leader in manufacturing processes with the introduction of 40nm lithography in 2008. It used 193nm immersion lithography as well as ultra-low connection material to improve chip performance and make it more. efficient.
These 40nm were divided into 2 processes: general use ( GP ) and low power ( LP ), which present a density 235% higher than the 65nm process. TSMC diversified this process to ULP as well, whose objective was to reduce consumption to the maximum. In the case of the GP, its main objectives were the following:
- CPU.
- GPU.
- Consoles.
- HDD.
- FPGA.
On the other hand, the 40nm LPs were focused on smartphones, DTT (DTV), decoders and wireless connectivity. As for the ULP process, its destination was laptops and the IoT.
2011-2013, TSMC achieves 28nm and 22nm process
Three years later, TSMC manages to manufacture the world's first 28nm technology, both for general use and for other options. There were 3 advantages of this process: more performance, more energy savings and more ecological. As in the previous process, those who benefited the most from this advance were:
- CPUs.
- GPUs.
- Network chips.
- Smartphones.
- Tablets.
- Automotive.
- IoT.
- Electronics.
The great novelty of this process is the use of the HMKG gate (High-k Metal Gate), whereas before a gate-first technology was used. The difference is that gate-last provides more advantages, among which less current leakage and better chip performance stand out.
The 22nm process, which came with Ultra Low Leak (ULL) technology, is worth mentioning. On the other hand, 22ULP came with a 10% reduced area, a speed gain of more than 30% and a consumption reduction that exceeded 30%. It was a process used by televisions, set-top boxes, etc.
2014, the fence tightens with 20nm
TSMC was once again the first manufacturer to advance the process with 20nm in 2014, but one of the keys was in the double pattern technology. As documented, TSMC's rise begins in 2015, when it surpassed 1 million shipments of 12-inch wafers.
Basically, we saw higher density and higher power than previous technologies, how? The quality of transistors and interconnections is improved, making them very efficient, as well as double pattern innovation. Compared to 28nm, the 20nm process achieves a 15% higher performance and reduces consumption by 1/3.
We have to say that, in 2013, TSMC was already preparing the 16nm and 12nm lithography. However, those technologies would take longer to come out.
2015, 16nm and 12nm processes arrive
In the processors still, we were on the 5th generation of processors Intel of 14nm, while AMD was by 28nm in their APU and chips FX of 32nm. First came the chips with 16 - nanometer process, being the first FinFET processor of 16nm fully functional in 2013.
Later, the so-called 16FF + (FinFet Plus) process would be introduced, which went into mass production in July 2015. One of TSMC's big customers for this process was the automotive industry, which demanded many chips of this type in 2017.
TSMC decided to further improve its 16nm FinFet technology, making it more profitable in 2016 with 16FFC. With this process, die production costs are cut by incorporating optical shrinkage and simplification simultaneously.
In 2017, we saw the 12nm FinFET process (12FFC) go into production. Thus, TSMC had a brutal process offering that served all types of customers:
- Many SoCs managed to double the power and consume 60% less.
- The real laptop boom begins.
2016, TSMC goes big with 10nm FinFET
We are in some very hectic years in computing, after seeing in December the presentation of the newest from AMD: Ryzen processors with Zen architecture. Although Intel did not know it, a future fierce fight was brewing in the desktop sector, and that is that the first AMD chips came with a 14nm process and with a Zen architecture that together offered enormous power.
The competition's biggest TSMC was GlobalFoundries, which manufactured the AMD processors. However, the Taiwanese went to them and did not stop working bringing the 10nm FinFET process to the market in 2016. The truth is that they would start manufacturing 10nm wafers on a large scale in 2017.
The news was the following:
- Aggressive geometric shrinkage.
- Double the logical density of 16nm chips.
- Speed increase of approximately 15%.
- A reduction in consumption of approximately 35%.
- Ideal for ASIC designs.
In 2018, Intel would launch its first x86 processors with a 10nm process, noting a major cutting edge difference with TSMC, which launched this process 2 years earlier.
2017-2019, 7nm puts TSMC on the map
Why this and not another? If we have to talk about TSMC's rise in attracting large clients, we must mention 7nm. It can be firmly stated that it is one of the best chip manufacturing processes ever made.
This process starts from a brutal innovation by TSMC, which makes it clear who the leader in the sector is. It was called N7 and was a process based on the FinFET transistor, offering 256MB SRAM with 2-digit yields. Mass production began in 2019, at which point the Ryzen 3000 and Zen 2 arrive - the first x86 processors to use 7nm.
TSMC was beginning to win over customers: Huawei, NVIDIA, AMD, Qualcomm, etc. More than 110 products were announced that carried the “N7” specification on their hardware, so TSMC continued to improve this process. So the 7nm FinFET Plus (N7 +) came out, which also went into production in 2019.
The flagship innovation was that it was the first ultraviolet EUV- capable manufacturing process technology. TSMC considers that this process is a solid basis for the 6nm FinFET technology that the Taiwanese company had already prepared in the same year, but it did not begin to produce the first samples until 2020.
If we compare the 10nm and 7nm processes we draw the following conclusions:
- Almost double the logical density.
- A speed gain of around 20%.
- 40% reduction in consumption.
We saw 2 different 7nm processes, at least on AMD:
- Zen 2 and the first 7nm process (N7).
- Zen 3 with a 7nm DUV or Deep Ultraviolet process.
2020-2021, Apple brings 5nm to the consumer market
Apple introduces the A14 Bionic as its first SoC of 5nm made history, but soon retaliate Samsung with its Exynos. AMD pointed in its roadmaps to 2021-2022 for the announcement of what would be the first 5nm x86 processors.
Well, this "Apple miracle" could not have been achieved without TSMC, and it is a process optimized for computer applications and high performance. It is the second extreme ultraviolet (EUV) lithography from the Taiwanese manufacturer, offering more density of transistors.
As usual, it brings reductions in energy consumption, the key to meeting the needs of 5G and AI, which attack the autonomy of the devices. The first samples began to be distributed in 2019, but volume production begins in the first half of 2020.
Differences between N7 and N5 :
- 15% speed improvement.
- Consumption reduction of 30%.
Finally, TSMC brings the novelty of N5P, a process that takes advantage of the design of N5, and that improves it in terms of its speed, consumption and size.
We hope this information has been helpful to you. If you have any questions, you can comment below and we will answer you shortly.
0 Comments