Germany Carl Zeiss, heart of Dutch ASML Lithography Equipment.

Status
Not open for further replies.

Anlsvrthng

Captain
Registered Member
For datacentres, but they still benefit from very large numbers of more advanced chips which use less electricity per calculation.

For mobile applications, limited battery life is the biggest issue. Cooling is secondary.
?
Have you seen the cooling system of a PC ?

Cooling is everything.

The performance improvement coming form architecture optimisation, NOT from the semiconductor manufacturing.

So, it come from clever redesign of circuit,not the manufacturing technologies of the above IC.
 

AndrewS

Brigadier
Registered Member
?
Have you seen the cooling system of a PC ?

Cooling is everything.

The performance improvement coming form architecture optimisation, NOT from the semiconductor manufacturing.

So, it come from clever redesign of circuit,not the manufacturing technologies of the above IC.

Desktops are dying out.
We see laptops/mobiles/datacentres being the main consumers of chips.

For datacentres, cooling is a function of how many electrons are dissipated as heat, which then requires more electricity for cooling.
For mobile applications, the amount of battery electricity is limited, and cooling is not really an issue.

So the key metric for all these applications is the amount of electricity used by the processor.
With smaller process nodes, smaller transistors use fewer electrons (and therefore electricity) to perform a calculation.

Yes, I agree architectural improvements are important.
But no matter what kind of architecture is used, they all benefit from a smaller process node which has smaller transistors, and therefore uses less electricity.

So a smaller process node has performance advantages, as long as you have large enough volumes of chips.
 

Anlsvrthng

Captain
Registered Member
Desktops are dying out.
We see laptops/mobiles/datacentres being the main consumers of chips.

For datacentres, cooling is a function of how many electrons are dissipated as heat, which then requires more electricity for cooling.
For mobile applications, the amount of battery electricity is limited, and cooling is not really an issue.

So the key metric for all these applications is the amount of electricity used by the processor.
With smaller process nodes, smaller transistors use fewer electrons (and therefore electricity) to perform a calculation.

Yes, I agree architectural improvements are important.
But no matter what kind of architecture is used, they all benefit from a smaller process node which has smaller transistors, and therefore uses less electricity.

So a smaller process node has performance advantages, as long as you have large enough volumes of chips.


Desktops dying out because there is no speed improvement.

There was a decade long learning curve in phones, and now they dying out as well, the sales flat/decreasing now.

There is no performance improvement in the last decade , so no one wants to buy desktop.

With phones there was a lot of money spent for phone capabilities, new screens/custom ICs and so on.

But it ended as well - there was no need to replace the desktop in the past 10 years, and now there is no need to replace the phone.

Incredible, you try to falsify me with data that actually support my point ?


In the case of data centres 70% of the new gen server CPUs contain ASIC circuits, tailored for specific customer.

They don't get the edge from the underlying technology - they managed to sell the products due to higher and higher customisation.
 

Anlsvrthng

Captain
Registered Member
45nm is not going to cut it for 5G phone , base station and AI chipsets when competitors are running at 7 or 5nm.
Because .... why?

It is a simple statement by you, with no data to support it.
If an IC will sell in 100 millions there is same (marginal) benefit to go for small node ( like slightly smaller mobile phone : D ) , but I seriously doubt that there will be any base-station controller on sub 22nm node.

The cost of a 7nm mask set/ tooling is in the 300 million ,means if you make one million IC then the tooling cost will be 300 dollars / IC .

For 5nm it is 550 million /IC.


Of course you are free to bring data falsify me : )

nano3.png
 

AndrewS

Brigadier
Registered Member
Desktops dying out because there is no speed improvement.

There was a decade long learning curve in phones, and now they dying out as well, the sales flat/decreasing now.

There is no performance improvement in the last decade , so no one wants to buy desktop.

With phones there was a lot of money spent for phone capabilities, new screens/custom ICs and so on.

But it ended as well - there was no need to replace the desktop in the past 10 years, and now there is no need to replace the phone.

Incredible, you try to falsify me with data that actually support my point ?


In the case of data centres 70% of the new gen server CPUs contain ASIC circuits, tailored for specific customer.

They don't get the edge from the underlying technology - they managed to sell the products due to higher and higher customisation.

No, desktops are dying out because they are big and bulky.
Laptops are small, portable and generally cheaper. They generally do have lower performance, but are perfectly fine for day-to-day use.
So the vast majority of consumers don't notice any performance difference between a desktop or laptop.
---
So why did you use a desktop analogy, if you already acknowledge that desktops are dying, and are being replaced by datacentres, laptops and smartphones?
---
Yes, ASICs are increasingly common for datacentres.
But remember ASICs still benefit from a smaller process node which uses less electricity per calculation.

Say you have a customised chip for a datacentre, and you have the choice of 14nm or 5nm.
Given high-enough production volumes, the 5nm chip ends up being the best option.

Why else are Bitmain and Canaan buying cutting-edge 5nm ASICs from TSMC?

If we use your logic on cooling constraints, Bitmain and Canaan should have stopped using smaller process nodes long ago.

Bitmain and Canaan to Reveal 5nm Bitcoin Mining Chips in 2020

Two of the largest bitcoin mining chip manufacturers have plans to equip their products with 5nm chips in the near future. Just recently, China-based manufacturer Bitmain purportedly produced the world’s first 5nm test-grade bitcoin mining chip and more efficient mining rigs will see mass production next year.
...
Following discussions concerning Bitmain’s upcoming ASIC rigs, regional reports revealed that Canaan will also have 5nm chips by the first quarter of next year. The 5nm chips will improve performance by 15% and the wafers will stem from the Taiwan-based semiconductor foundry.

Please, Log in or Register to view URLs content!
 

nlalyst

Junior Member
Registered Member
45nm is not going to cut it for 5G phone , base station and AI chipsets when competitors are running at 7 or 5nm.
I think it is important to differentiate the digital from analog components. While the digital part of the state-of-the-art 5G modems is manufactured at 7nm, the analog RFIC are manufactured at 28nm. These analog chips are made using a SOI process, with GlobalFoundries and TowerJazz being one of the industry leaders and Soitec of France being the largest supplier of RF SOI substrates.
 
Status
Not open for further replies.
Top