You may still have 20nm and 10nm generations | of manufacturing processes left behind by the desktop. In both cases the problem was the same – the shift in the parameters essential to the desktop would be weak, but the price shift did not match. The former was originally intended to be in two versions – mobile and powerful, but the executive eventually canceled TSMC on the grounds that the mobile option had achieved the parameters planned for the executive, so it would not be needed. The real reason was that the planar process without FinFET ran into the limits of its capabilities (accessible with the technologies at that time) and the executive version failed – it offered no more than a mobile one. The 10nm process has only been mobile.
In the case of the 7nm generation, the TSMC has prepared both the mobile variant | so powerful, so it was expected that with 5 nanometers it would only remain mobile. Eventually it looks different. TSMC also plans a powerful 5nm process variant suitable for processors and graphics chips. She would not prepare it here if she did not feel the demand for such a combination. In other words, at least one of the manufacturers is preparing 5nm x86 processors or separate graphics cards.
What processors could be | created at the 5th TSMC process? Zen is 14nm, Zen + 12nm, Zen 2 7nm and from Zen 3 expected 7nm + or 7nm with EUV. Since the AMD roadmap was created in 2020, a refresh Zen 2+ on the 7nm process could theoretically arise. That would be followed in 2021 by Zen 3 at 7 nanometers, and then there would be a possibility of using a 6nm process or 5nm. The 6nm process brings a milder shift, but is designed to move from 7nm designs that can be a little bit smaller and a little faster. 5nm no longer offers this option, requires a process-tailored design, but increases the density by 80%, which means that the same design area will be reduced by about 40% compared to the 7nm process by using 5nm.
|7nm LPP ( 2nd gene)||–||7nm (1st gene)||–|
|7nm (3rd gene) = ( ?) 6nm||+ 17% vs. 7LPP||7nm + (EUV / 2nd gen)||+17% vs. | 7nm|
|5nm||+ 25% vs. | 7LPP||6nm||+18% vs. | 7nm|
|3 / 4nm MBCF||“full node”||5nm||7nm|
So there would be a theoretical possibility | 7nm EUV Zen 3 year after release to shrink to 6nm process at Zen 3+ ( 2022) and in 2023 come with 5nm Zen 4 . In view of the fact that an iPhone with a 5nm chip is due to be released in 2020 and only three quarters of a year have passed between the first 7nm iPhone and the first 7nm x86 processor, the release of the first 5nm processor in 2023 seems quite conservative. So it is possible that Zen 5 would arrive a year earlier in the 5th process (if the evolution of architecture allows it), or some variation will be passed on to this process Zen 3+ (a scenario where separate processors can be imagined) Zen 3+ will be produced on a more powerful 5nm process and Zen 3+ APU on a more affordable 6nm
In a GPU world, the situation will be more complicated because (especially high-end GPUs can not respond flexibly to new processes, yielding many hundreds of millimeters of chips typical of more powerful graphics chips is significantly lower and therefore usually longer than the process is ready. for desktop chips and only TSMC will offer three alternatives: 7nm + (7nm with EUV), 6nm and 5nm.
Samsung will bring more options with its third generation 7nm EU In the process, with the 5nm process (which will be closer to the 6nm by TSMC), the new 3nm BMCFET technology, which is optimistic about the end of 2021 in the form of Samsung serial production, but above which a number of question marks are hanging. likely to move to later years. At least in terms of desktop usage.