Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/what-were-the-historical-drivers-of-foundry-node-investments-for-tsmc.17408/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021370
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

What were the historical drivers of Foundry node investments for TSMC?

Xebec

Well-known member
I know there's likely no real simple answer here, but I'm curious any way.

I was just thinking that for Intel, (oversimplified) the driver for foundry progress since the early 1980s has always been Intel CPUs (and chipsets).. and a mixture of CPUs and Intel memory prior to that. (+FPGA also driving some of this).

What did the journey for TSMC look like?

For example, Apple's SoC demand followed by GPUs, CPUs, and AI/ML accelerators are driving the latest nodes (sub-10nm) at TSMC. In broad swaths, what did the earlier eras look like for TSMC? Say if we broken into:

28nm to 10nm: Apple, dGPUs, FPGAs?
65nm to 28nm: ???
180nm to 65nm:
etc.

I'm really curious about these older nodes as my primary exposure to TSMC prior to 28nm was always wondering about the next GPU, and not paying much attention to the other markets.

Thanks!
 
Before smartphones, there were cell phones, with Nokia and Motorola making modems for 2G and 3G digital phones. Before that there were analog modems. Wifi became big in the early 2000s, those chips would have been made at foundries.

There have also been non-X86 CPUs, like SPARC, Transmeta, Motorola/Freescale iMX and PowerPC, NEC logic chips, STMicro, Infineon. These remain fab lite applications, where older logic chips are made in house, down to 90nm or so. That is pretty much where 200mm fabs end. TSMC picked up a lot of business from the "applications too small for a wholly-owned 300mm fab" crowd.

FPGAs are another significant application that was from the beginning fabless. Altera and Xilinx, from the beginning.
 
I believe Altera (FPGA) and Nvidia (GPU) led TSMC's leading edge investment. If memory serves me right there was a time when NVDA was ~20%+ revenue customer - early 2000 as per my recollection. A large NVDA order turned around TSMC's revenue/ utilisation post dot com bubble decline
 
I believe Altera (FPGA) and Nvidia (GPU) led TSMC's leading edge investment. If memory serves me right there was a time when NVDA was ~20%+ revenue customer - early 2000 as per my recollection. A large NVDA order turned around TSMC's revenue/ utilisation post dot com bubble decline

Altera was a very close TSMC partner from the very beginning of both companies. I worked with Altera a bit way back when. TSMC used Altera for process development and pipe cleaning until of course Xilinx came to TSMC at 28nm. TSMC treated Altera and Xilinx equally and Xilinx wiped the floor with Altera. Altera went to Intel and the rest is history. Xilinx was one of the first fabless companies, interesting history. We talked about them in our book:


I think QCOM was TSMC's leading customer back then even though they whore'd around quite a bit. QCOM would develop SoCs with TSMC then 2nd source to others, up until Apple of course then QCOM fled to Samsung 14nm but they are back at TSMC at N5-3. Now QCOM is working closely with IFS. A rolling stone gathers no moss.
 
I know there's likely no real simple answer here, but I'm curious any way.

I was just thinking that for Intel, (oversimplified) the driver for foundry progress since the early 1980s has always been Intel CPUs (and chipsets).. and a mixture of CPUs and Intel memory prior to that. (+FPGA also driving some of this).

What did the journey for TSMC look like?

For example, Apple's SoC demand followed by GPUs, CPUs, and AI/ML accelerators are driving the latest nodes (sub-10nm) at TSMC. In broad swaths, what did the earlier eras look like for TSMC? Say if we broken into:

28nm to 10nm: Apple, dGPUs, FPGAs?
65nm to 28nm: ???
180nm to 65nm:
etc.

I'm really curious about these older nodes as my primary exposure to TSMC prior to 28nm was always wondering about the next GPU, and not paying much attention to the other markets.

Thanks!

There are slides on this. When I joined the industry in the early 1980s it was computers. Mainframes then minicomputers (IBM, HP, DEC, Prime, DG, etc...) Most computer companies had their own fabs and made their own chips. Then came the IDMs Intel/Motorola and the desktop PC revolution started followed by laptops then mobile devices like pagers and early mobile phones. I had one of the first Motorola car phones that were installed in the car. Then came flip phones and Blackberrys and of course smart phones. Today HPC/AI drive the semiconductor industry and I think that will continue.
 
I think QCOM was TSMC's leading customer back then even though they whore'd around quite a bit. QCOM would develop SoCs with TSMC then 2nd source to others, up until Apple of course then QCOM fled to Samsung 14nm but they are back at TSMC at N5-3. Now QCOM is working closely with IFS. A rolling stone gathers no moss.
On the bright side it gives us a great way to benchmark different process technologies. Although with the NUVIA buy that might get harder as QCOM diverges from off the shelf ARM cores.
 
Back
Top