Apple, TSMC and other chip giants, join forces to create "nano lies"?
【Text/Observer Network Lv Dong】
"If 4nm is not really 4nm, then does the node name still make sense?" "If an older process can be renamed, why incur the extra cost of 4nm technology?" ”
When Apple launched the iPhone 14 series last year, the biggest adjustment made compared to the past was to carry the latest 4nm A16 processor on the iPhone 14 Pro series, and the ordinary version used the previous generation of 5nm A15 processor.
At that time, the market generally believed that Apple did this to open the gap between the Pro series and the ordinary version. In this way, on the one hand, the high price of the Pro series can be maintained, and on the other hand, the sales of the Pro series can be stimulated with new processors.
However, the foreign technology media "Wccftech" reported on August 21 that Apple's marketing team is suspected of misleading consumers, because Apple's 4nm A16 processor is internally marked as a 5nm processor. The source cited in the report was Twitter user @URedditor, who was called a "well-known whistleblower" by "Wccftech."
In fact, before the birth of Apple's A16 processor, there were already professional institutions questioning the advanced processes of TSMC and Samsung.
For example, market research firm TechInsights published a study last year titled "Lies About Nano" that two leading foundries allowed customers to claim that they used the 4nm process, but they were still using 5nm technology. This situation has tarnished the image of both parties, especially the foundries. Behind this, it also means that the development of transistors is slow.
Source: TechInsights
A16 is basically the same as A15?
Apple is not @URedditor more than questionable.
Last October, just a month after the release of the iPhone 14 series, Jason Cross, a longtime Apple-following "Macworld" analyst, dug into the A16 processor and thought it was basically the same as the A15.
According to Apple, the A16 processor is manufactured using TSMC's latest 4-nanometer process. However, it is worth noting that TSMC's 'N4' is not a 4nm process in the true sense. While the A16 is more advanced than earlier A-series processors, it doesn't use a true next-generation manufacturing process. Cross said.
Source: 9to5mac
Below this comment, there is also a follow-up post that points out:
"The A16 is an incremental improvement over the A15, which explains why the regular iPhone 14 can get features such as action mode, photonic engine and 4K movie mode. It makes it clearer than ever that Apple always installs software lock features on older devices, especially the iPhone 13 series, which can be easily obtained. ”
According to Cross's research, the basic structure of the A16 is the same as that of the A15: it has two high-performance cores, four efficiency cores, five GPU cores, and 16 neural engine cores. The number of transistors has increased, but only from 15 billion to 16 billion, an increase much smaller than usual.
The news cited by "Wccftech" said that Apple's upcoming A17 processor will use TSMC's 3nm process, and single-core and multi-core performance will be increased by 31%. Apple seems to be going to adopt last year's strategy and carry the A17 on the iPhone 15 Pro series, while the regular version still uses the A16.
"Of course, we have to get to know if the A17 processor is just another product that Apple is fully marketing, or a chip designed and mass-produced on an entirely new architecture." We are disappointed that the A16 is not 4 nanometers, but this is not surprising as businesses always follow this practice. The report said.
Source: Wccftech
"Nano's Lie"
In fact, before the birth of Apple's A16, market research agency TechInsights issued a report questioning TSMC and Samsung's 4 nanometers.
According to the report, in a long-term competition with TSMC, Samsung announced that it will deliver production of 5-nanometer chips by the end of 2021, a year after delivering 4-nanometer chips. At that time, TSMC planned to deliver 4nm in the second quarter of 2022 in two years between the 5nm and 4nm nodes.
But later, in order to avoid giving Samsung the opportunity to "show off its might", TSMC decided to "accelerate" the progress of its N4 nodes by two quarters to catch up with competitors.
Source: TechInsights
At the end of 2021, the first MediaTek Dimensity 9000 using TSMC's 4nm process was released.
"TSMC should have been skeptical that it could suddenly squeeze six months out of its usually rigorous and time-consuming production certification cycle." TechInsights said.
When analyzing the Dimensity 9000, TechInsights found that the key process size was exactly the same as TSMC's earlier 5nm product, "the situation is not good, the foundry's claim of 4nm product is false, just as MediaTek claims to have a 4nm processor."
At the same time, Samsung's smartphone division is preparing to launch the Galaxy S22, which is equipped with Samsung's self-developed 4nm processor Exynos 2200 and Qualcomm's 4nm processor Snapdragon 8 Gen 1. To produce both products, Samsung has also created a new process called 4LPX.
But TechInsights, after analyzing the Snapdragon 8 Gen 1, found that 4LPX is physically no different from 5LPE.
What's even more embarrassing is that Samsung's later products that actually use 4nm technology perform worse than "fake" 4nm products. As Samsung was too aggressive in putting its 4LPE process into production, resulting in the Exynos 2200's high defect rate and low energy efficiency, these difficulties prompted Qualcomm to turn to TSMC.
According to the TechInsights report, the Qualcomm Snapdragon 8+ Gen 1 released in May 2022 uses TSMC's updated version of N4P, "which we believe is the real 4nm technology."
At the time, the research agency also expected that the Apple A16 would also use TSMC's "true 4-nanometer technology" - N4P.
However, according to the aforementioned revelations, the Apple A16 unveiled in September 2022 may still use N4 instead of the updated version of N4P.
Apple A16 processor
"Meaningless node name"
"Why are foundries suddenly misreporting their transistor progress? It's much easier than acknowledging weak progress." ”
TechInsights reports that the difference between a fake 4nm process and a "real" 4nm process is only 5% optical shrinkage (10% area reduction). But even that small advance shattered Samsung's production model, taking TSMC two years to complete.
Given the long interval between nodes, foundries may increase speed, power, or yield (and ultimately achieve the same effect) while waiting for the next node. For example, while 4LPX has the same dimensions as 5LPE, the former may offer some additional advantages to justify its deeds to a new name.
But TechInsights believes that in the absence of any increase in transistor density, the process name should be 5LPX to indicate that the process still falls under the 5nm category. Because most people replace the full process name with a 5nm or 4nm abbreviation, numbers are still important.
"Although counterfeiting new nodes can save foundries from short-term difficulties, it is harmful to them."
The report questioned that touting a "fake" 4nm milestone and allowing customers to hype it would dilute the value of the foundry brand. If 4nm is not really 4nm, then does the node name still make sense? If an older process could be renamed, why incur the extra cost of 4nm technology? ”
While chip foundries haven't maintained a strict correlation between node names and transistor densities for some time, at least when they adopt a smaller nanonumber in their name, it shows that they have made some progress in density.
TechInsights bluntly stated that blatantly relabeling a 5nm process as 4nm erases the last shred of significance of node numbers.
"So when a foundry claims that it has reached the next level, don't believe it until the disassembly is confirmed." The same applies to chip suppliers who boast of implementing new manufacturing technologies themselves. As Moore's Law stumbles to an end, we are left with lies, damn lies, and meaningless node names. The report reads.