Link?
Chinese semiconductor industry
- Thread starter Hendrik_2000
- Start date
- Status
Dongguan Pasa Electronic Equipment Co., Ltd. is a professional semiconductor process equipment manufacturer integrating development, production and sales. It is a key backbone enterprise of China's high-end electronic industry basic equipment.
The company's main products are: special process equipment for semiconductor technology, special process equipment for electronic components, integrated circuit process equipment, special process equipment for solar cells, LED process equipment, gallium arsenide process equipment, nanomaterial process equipment, magnetic material process equipment, There are more than 100 types of special customized process equipment.
The company's main products are: special process equipment for semiconductor technology, special process equipment for electronic components, integrated circuit process equipment, special process equipment for solar cells, LED process equipment, gallium arsenide process equipment, nanomaterial process equipment, magnetic material process equipment, There are more than 100 types of special customized process equipment.
Companies that sell semiconductor-nanotech tools for R&D and possibly customized small batch production.
The also sell what looks like a really small sized Ion Implanter.
Low-energy ion implanter
The low-energy ion implanter can generate ultra-low-energy ion beams through the ion source. After being deflected by a dipole magnet, ions with different charge-to-mass ratios are filtered out, leaving only the required ion types and energies
Even more on this
This really does seems like a very interesting approach. They first tried this computation in SRAM approach last year on a 22nm process. Now, they are able to generate even more powerful & efficient chip with 12nm process
so there is a well known issue where 60% of time is spent on data transfer and over 90% of power is used in data transfer. So my putting the computation unit inside SRAM (it's not clear how they are doing this), but they claim to greatly reduce # of calls into memory unit and computation unit, making it a lot more efficient. Didn't mission hybrid bonding or anything like this
It's known that computation speed is advancing faster than memory bandwidth. By placing them close together, this just decreases the stale time of the computation unit. Alibaba Damo academy put memory/computation in 1 unit as a top 10 trend in 2023.
Looks like this has been supported by 55 possible customers. From what I can see, this is not limited to autonomous driving. If they scale this up and make it on 7nm process, this could be a very powerful GPGPU
It seems like this is a very new concept, so there is a lot of improvement ahead still
before this instrument can be used as a production tool, we have to think about some practical things:
1. what about the absorption cross section of 6.7 nm vs 13.5 nm in typical photoresist?
2. what is the efficiency of the Gd vs Sn source?
3. what is the reflective efficiency of internal mirrors at 6.7 nm vs 13.5 nm?
yes, when it is done, it should halve the CD size, but even then CD is not purely dependent on wavelength. CD = k*L/NA where L is wavelength, NA is numerical aperture, k is a parameter of the system.
Status of various SMIC new fabs based on Q1 report. They will maintain capex.
Shenzhen has started mass production
Jingcheng to enter mass product in 2nd half
Lingang to open line at end of this yr
Tianjin is still under construction
So it seems to me that Jingcheng has taken the longest to get up an running, because it's taken a lot of effort to test out de-americanized production lines. Hopefully, Lingang come online a lot faster.
Q1 capex was only 1.26B
capex maintaining earlier projection (last year's level), so expect Q2 capex to be high imo as it takes delivery of more ASML scanners before sanctions hit. Works out to be $5B for rest of the yr
the utilization dropped down to an unsightly 68%
monthly capacity increased to 732k 8-inch equivalent from 714k in Q4 and 649k a yr ago. Not the largest jump, but I think it's due a combination of already low utilization + delays in Jingcheng plant
They sold 1.251m wafers in Q1 vs 1.574m in Q4 and 1.84m a yr ago(converted to 8-inch wafer, so keep that in mind)
Revenue was $1.46B in Q1 vs 1.62B in Q4 and 1.84B a yr ago
That means revenue per 8-inch wafer has gone up from
$1000/wafer a year ago to $1029/wafer in Q4 to $1167/wafer in Q1
Since I doubt the price per wafer has really gone up. More likely went down. I think this is likely due to two factors:
1) Higher Finfet sales as % of overall sales -> its Finfet production likely fully booked and slowly increasing in capacity
2) Achieving higher wafer prices on more advanced process like 12/14nm through getting higher yield and more complexity on them
thoughts?
Last edited:
Any idea on the nodes in the facilities?
28nm. De-Americanized line.
Your theory on higher FinFET sales resulting in higher ASP makes sense. But that’s not it.
There were several questions regarding the ASP during the investor call. The answer from SMIC CFO(?) was that it’s partially due to lower 8” order/utilization and higher price from rush orders for new products in 12”.
I recommend you listen to the call…(1hr). They also talked about their strategy on what product to focus on for their new fabs.
- Status