TSMC has cheaper 4nm node, making low-end CPUs, GPUs viable

N4C: Low-cost version of 4nm technology

State-of-the-art process nodes are getting more expensive, and even the cost per transistor isn’t dropping anymore, so we’ll increasingly see cheap CPUs and GPUs being unable to use the latest processes. Fortunately, TSMC is also preparing lower-cost versions alongside the high-end nodes. Right now, 6nm node serves the role, but cheaper 4nm manufacturing should soon be available instead, which could help a lot especially for GPUs.

We’ve already reported on TSMC’s unveiling of the future 1.6nm A16 manufacturing node for the most advanced chips at the North American Technology Symposium 2024, but it’s likely to be quite expensive. But with it, the company also presented the N4C process, which in turn is aimed at lowering the price.

Read more: TSMC announced a 1.6nm node. First with Backside Power Delivery

It is a node derived from the 5nm generation (N5), so when we say cheaper, it should be seen relative to the overall high cost level of these EUV process nodes – the production will still be a comparatively expensive affair. The N4C node is supposed to be an extension of the high-performance N4P technology, but with additional modifications to reduce the cost of manufacturing a particular chip design, where TSMC says the savings can be as high as 8.5%. Compared to using the 3nm node, which has a significantly higher cost than the N5 and N4P (and now N4C) generations, the price will of course be even lower.

A major factor in this cost reduction is the reduction in the number of masks the node uses and the related number of patterning steps – fewer steps in wafer processing means more production line capacity and therefore lower cost. TSMC has also changed some chip design rules for the N4C node, and the node provides redesigned standard blocks and SRAM libraries. Their modifications are likely aimed at reducing the area of these blocks and thus reducing the die area (and thus the price) of the designed chip.

Thus, the N4C node is supposed to achieve a higher density of transistors or smaller die area, which, in addition to other cost-cutting changes, can also have an indirect effect on the economics of production, because for the same number of defects on the wafer there is (due to the fact that there are more chips on the wafer), you will end up with smaller relative percentage of defective dies, if there is more of them on the wafer.

The economical 4nm N4C node (Author: TSMC, via: AnandTech)

Unfortunately, TSMC does not state that power efficiency will improve, so it will probably remain no better than for N4P. And there is no mention of performance either. However, libraries optimized for smaller die area and price could lead to lowered performance or clocks. It is therefore possible that on N4C, for example, CPU designs will not achieve as high clock speeds as on the more powerful/expensive N4P version of the 4nm node.

Production chips in 2025–2026

TSMC states that mass production with this technology will begin in 2025. Typically, it takes several months for any products to actually appear on the market, so in practice we will probably see such chips in 2026 (but 2025 is not out of question, if the start of mass production is somewhere near the beginning of the year rather than later in).

This technology could be used to produce lower price segment processors for both mobile phones and PCs, but also cheaper GPUs, SSD controllers and also perhaps some less performance-critical chipsets in CPUs and GPUs that otherwise might use the latest 3nm, 2nm and even smaller technologies. This is likely to be a technology that will be used quite widely and long after the more advanced Angstrom era nodes arrive.

TSMC nodes roadmap (Author: TSMC, via: AnandTech)

The low price of N4C will be relative

As mentioned, it is important to remember that the price reduction is based on 5nm EUV technology, which is itself quite expensive. Despite the availability of the cheaper N4C variant, this technology may still not be exactly affordable and may still not be well applicable to the very cheapest products (e.g. processors you see in the cheapest laptops).

Read more: AMD to produce lowend CPUs and GPUs using Samsung’s 4nm node

According to reports that have been leaking for some time, AMD is going to manufacture some products at Samsung precisely due to these factors. Prices offered by the Korean foundry could probably be even lower than what you’ll pay for TSMC’s N4C node.

The company reportedly plans to use Samsung’s technology for the Sonoma Valley processor (or SoC), which will be a more modern equivalent of the low-end Mendocino APU, this time based on the Zen 5c architecture. It is possible that Intel will similarly try to lure in customers using bargain pricing, and the pressure of this competition may potentially lead to lower prices of TSMC’s services too, which may include the N4C node.

Source: AnandTech

English translation and edit by Jozef Dudáš


  •  
  •  
  •  
Flattr this!

Leave a Reply

Your email address will not be published. Required fields are marked *