Vertical Divider
Intel’s Program to Recapture IC Leadership
Intel's roadmap has recovered last year's 7nm delay: the node that was previously called 5nm has seen pretty much no cascading delays. That node was the one with which Intel (in 2019) intended to recapture process. In March, Intel’s new CEO Pat Gelsinger shook announced that the Intel Foundry Services would compete against Samsung and TSMC. In July 26, Intel reported that despite the 7nm delay, the revolutionary 20A node (previously called 5nm) has not been delayed. As Intel plans to regain process leadership with (stacked) RibbonFET, PowerVia, and high-NA EUV, they announced new, significant wins with AWS and Qualcomm. After the 7nm delay last year, many thought Intel had become a dinosaur that would irreversibly fall behind. Instead, TSMCs current leadership could be finished by 2024-2025.
After the 7nm delay last year, many thought Intel's days were numbered. Instead, Intel quickly punched back and embraced EUV to fix 7nm in record time. Although Pat Gelsinger said Intel will deliver 5 nodes in four years, the time delta between the second and the fifth node actually seems closer to just two years (2023-2025), since there are only two full-node transistor shrinks, which were previously called 7nm and 5nm, each with one further intra-node improvement). The simplest way to look at the Intel Accelerated announcements is that Intel has been able to mitigate the waterfall effect of the mid-2020 7nm (now called Intel 4) delay and would closely match the 2029 roadmap Intel provided in late 2019:/ Just as significantly is Intel's packaging roadmap, with two new announcements. Intel Foveros Omni and Direct will both launch in 2023.
Foveros Omni will be for 3D packaging what EMIB brought for 2.5D packaging. Namely, EMIB solved the problem of having to use large, expensive interposers to connect chips in 2D. From how Intel described Omni, it seems this technology will accomplish exactly the same for the Foveros 3D packaging family. As the name implies, Omni supports both horizontal as well as vertical interconnections, giving chip designer extreme, unmatched flexibility, at low cost.
For example, I have been arguing for over a year that there will be no N3-based iPhones in 2022, despite many thinking/assuming that will be the case. This is because although TSMC announced N3 for 22H2 volume production, there is simply no way to start volume production and get an iPhone launch in September or October 2022 - even if N3 production started on July 1, 2022. This is simply because manufacturing a chip takes months.
Hence, when Intel says for example that Intel 20A will be ready in 24H1, it could be the case that there will be no products until 24H2. The reason it is unclear if it is referring to production or product launches is because Intel previously announced that Meteor Lake would launch in "early 2023” and has since remained committed (including at the Accelerated event) that 7nm (now called Intel 4) will be ready in 23H1.
Intel's roadmap has recovered last year's 7nm delay: the node that was previously called 5nm has seen pretty much no cascading delays. That node was the one with which Intel (in 2019) intended to recapture process. In March, Intel’s new CEO Pat Gelsinger shook announced that the Intel Foundry Services would compete against Samsung and TSMC. In July 26, Intel reported that despite the 7nm delay, the revolutionary 20A node (previously called 5nm) has not been delayed. As Intel plans to regain process leadership with (stacked) RibbonFET, PowerVia, and high-NA EUV, they announced new, significant wins with AWS and Qualcomm. After the 7nm delay last year, many thought Intel had become a dinosaur that would irreversibly fall behind. Instead, TSMCs current leadership could be finished by 2024-2025.
After the 7nm delay last year, many thought Intel's days were numbered. Instead, Intel quickly punched back and embraced EUV to fix 7nm in record time. Although Pat Gelsinger said Intel will deliver 5 nodes in four years, the time delta between the second and the fifth node actually seems closer to just two years (2023-2025), since there are only two full-node transistor shrinks, which were previously called 7nm and 5nm, each with one further intra-node improvement). The simplest way to look at the Intel Accelerated announcements is that Intel has been able to mitigate the waterfall effect of the mid-2020 7nm (now called Intel 4) delay and would closely match the 2029 roadmap Intel provided in late 2019:/ Just as significantly is Intel's packaging roadmap, with two new announcements. Intel Foveros Omni and Direct will both launch in 2023.
Foveros Omni will be for 3D packaging what EMIB brought for 2.5D packaging. Namely, EMIB solved the problem of having to use large, expensive interposers to connect chips in 2D. From how Intel described Omni, it seems this technology will accomplish exactly the same for the Foveros 3D packaging family. As the name implies, Omni supports both horizontal as well as vertical interconnections, giving chip designer extreme, unmatched flexibility, at low cost.
For example, I have been arguing for over a year that there will be no N3-based iPhones in 2022, despite many thinking/assuming that will be the case. This is because although TSMC announced N3 for 22H2 volume production, there is simply no way to start volume production and get an iPhone launch in September or October 2022 - even if N3 production started on July 1, 2022. This is simply because manufacturing a chip takes months.
Hence, when Intel says for example that Intel 20A will be ready in 24H1, it could be the case that there will be no products until 24H2. The reason it is unclear if it is referring to production or product launches is because Intel previously announced that Meteor Lake would launch in "early 2023” and has since remained committed (including at the Accelerated event) that 7nm (now called Intel 4) will be ready in 23H1.
Intel's roadmap has recovered last year's 7nm delay: the node that was previously called 5nm has seen pretty much no cascading delays. That node was the one with which Intel (in 2019) intended to recapture process. In March, Intel’s new CEO Pat Gelsinger shook announced that the Intel Foundry Services would compete against Samsung and TSMC. In July 26, Intel reported that despite the 7nm delay, the revolutionary 20A node (previously called 5nm) has not been delayed. As Intel plans to regain process leadership with (stacked) RibbonFET, PowerVia, and high-NA EUV, they announced new, significant wins with AWS and Qualcomm. After the 7nm delay last year, many thought Intel had become a dinosaur that would irreversibly fall behind. Instead, TSMCs current leadership could be finished by 2024-2025.
After the 7nm delay last year, many thought Intel's days were numbered. Instead, Intel quickly punched back and embraced EUV to fix 7nm in record time. Although Pat Gelsinger said Intel will deliver 5 nodes in four years, the time delta between the second and the fifth node actually seems closer to just two years (2023-2025), since there are only two full-node transistor shrinks, which were previously called 7nm and 5nm, each with one further intra-node improvement). The simplest way to look at the Intel Accelerated announcements is that Intel has been able to mitigate the waterfall effect of the mid-2020 7nm (now called Intel 4) delay and would closely match the 2029 roadmap Intel provided in late 2019:/ Just as significantly is Intel's packaging roadmap, with two new announcements. Intel Foveros Omni and Direct will both launch in 2023.
- Foveros Direct should be regarded as Intel's equivalent to TSMC's Wafer-on-Chip since both technologies use hybrid bonding with sub-10um bump pitches. Although AMD (NASDAQ:AMD) will already commercialize hybrid bonding in 2022 with its 3D V-Cache, AMD has explicitly said the product would be limited to high-end desktop CPUs, while Intel said during the Q&A that it is looking to commercialize the product first on its mainstream mobile CPUs. Note that Intel quite recently announced a $3.5B investment to produce Foveros at Rio Rancho; Foveros will be a high-volume technology by 2023.
- Although hybrid bonding (whether from Intel or TSMC) is getting most attention due to the 3D V-Cache announcement, I am actually most bullish about Foveros Omni, which leverages a truly unique Intel packaging technology.
Foveros Omni will be for 3D packaging what EMIB brought for 2.5D packaging. Namely, EMIB solved the problem of having to use large, expensive interposers to connect chips in 2D. From how Intel described Omni, it seems this technology will accomplish exactly the same for the Foveros 3D packaging family. As the name implies, Omni supports both horizontal as well as vertical interconnections, giving chip designer extreme, unmatched flexibility, at low cost.
For example, I have been arguing for over a year that there will be no N3-based iPhones in 2022, despite many thinking/assuming that will be the case. This is because although TSMC announced N3 for 22H2 volume production, there is simply no way to start volume production and get an iPhone launch in September or October 2022 - even if N3 production started on July 1, 2022. This is simply because manufacturing a chip takes months.
Hence, when Intel says for example that Intel 20A will be ready in 24H1, it could be the case that there will be no products until 24H2. The reason it is unclear if it is referring to production or product launches is because Intel previously announced that Meteor Lake would launch in "early 2023” and has since remained committed (including at the Accelerated event) that 7nm (now called Intel 4) will be ready in 23H1.
Intel's roadmap has recovered last year's 7nm delay: the node that was previously called 5nm has seen pretty much no cascading delays. That node was the one with which Intel (in 2019) intended to recapture process. In March, Intel’s new CEO Pat Gelsinger shook announced that the Intel Foundry Services would compete against Samsung and TSMC. In July 26, Intel reported that despite the 7nm delay, the revolutionary 20A node (previously called 5nm) has not been delayed. As Intel plans to regain process leadership with (stacked) RibbonFET, PowerVia, and high-NA EUV, they announced new, significant wins with AWS and Qualcomm. After the 7nm delay last year, many thought Intel had become a dinosaur that would irreversibly fall behind. Instead, TSMCs current leadership could be finished by 2024-2025.
After the 7nm delay last year, many thought Intel's days were numbered. Instead, Intel quickly punched back and embraced EUV to fix 7nm in record time. Although Pat Gelsinger said Intel will deliver 5 nodes in four years, the time delta between the second and the fifth node actually seems closer to just two years (2023-2025), since there are only two full-node transistor shrinks, which were previously called 7nm and 5nm, each with one further intra-node improvement). The simplest way to look at the Intel Accelerated announcements is that Intel has been able to mitigate the waterfall effect of the mid-2020 7nm (now called Intel 4) delay and would closely match the 2029 roadmap Intel provided in late 2019:/ Just as significantly is Intel's packaging roadmap, with two new announcements. Intel Foveros Omni and Direct will both launch in 2023.
- Foveros Direct should be regarded as Intel's equivalent to TSMC's Wafer-on-Chip since both technologies use hybrid bonding with sub-10um bump pitches. Although AMD (NASDAQ:AMD) will already commercialize hybrid bonding in 2022 with its 3D V-Cache, AMD has explicitly said the product would be limited to high-end desktop CPUs, while Intel said during the Q&A that it is looking to commercialize the product first on its mainstream mobile CPUs. Note that Intel quite recently announced a $3.5B investment to produce Foveros at Rio Rancho; Foveros will be a high-volume technology by 2023.
- Although hybrid bonding (whether from Intel or TSMC) is getting most attention due to the 3D V-Cache announcement, I am actually most bullish about Foveros Omni, which leverages a truly unique Intel packaging technology.
Foveros Omni will be for 3D packaging what EMIB brought for 2.5D packaging. Namely, EMIB solved the problem of having to use large, expensive interposers to connect chips in 2D. From how Intel described Omni, it seems this technology will accomplish exactly the same for the Foveros 3D packaging family. As the name implies, Omni supports both horizontal as well as vertical interconnections, giving chip designer extreme, unmatched flexibility, at low cost.
For example, I have been arguing for over a year that there will be no N3-based iPhones in 2022, despite many thinking/assuming that will be the case. This is because although TSMC announced N3 for 22H2 volume production, there is simply no way to start volume production and get an iPhone launch in September or October 2022 - even if N3 production started on July 1, 2022. This is simply because manufacturing a chip takes months.
Hence, when Intel says for example that Intel 20A will be ready in 24H1, it could be the case that there will be no products until 24H2. The reason it is unclear if it is referring to production or product launches is because Intel previously announced that Meteor Lake would launch in "early 2023” and has since remained committed (including at the Accelerated event) that 7nm (now called Intel 4) will be ready in 23H1.
|
Contact Us
|
Barry Young
|