- Dec 25, 2013
- 3,899
- 193
- 106
Note: I know the title is similar to a recent article by AnandTech, and due to some recent feedback, I should mention that the content is markedly different, though. Read the AnandTech article if you want just info about 2019's CPUs. I know 2020 already started, but might still be of some interest.
Intel 2019 Year in Review
For Intel, 2019 marked an important, albeit turbulent, year. After several years of stagnation on the technology side, the company moved forward with a much-improved architecture on its ‘new’ 10nm process. Other introductions included its Nervana artificial intelligence push, 96-layer 3D NAND and more.
Several storylines also unfolded throughout the year, such as Intel exiting the 5G smartphone modem space, the company’s ongoing struggles with shortages, its plans to return to an aggressive two-year process cadence after 10nm’s three year delay and also plans for a global robotaxi business as part of Mobileye.
Shortages and 5G
These storylines started began in 2018 already. The year started in the wake of the 14nm supply issues, which have continued to persist throughout the year. Intel’s explanation was a combination of unprecedented growth in the data center and a better than expected PC market, the delays of 10nm that resulted in larger 14nm dies to improve performance, and the transition of its chipsets and 4G modems to 14nm and increased modem share.
In response, in December 2018 Intel announced major, multi-dollar factory expansion plans, planned a new $11 billion fab in Israel within the next four years, and progressed on equipping the $9 billion Fab 42 in Arizona for 7nm. When it finally starts production in 2021, it will have been ten years since the fab was announced in 2011. More recently, Intel said it has increased wafer capacity by 25% in 2019 and plans another 25% increase in 2020. In Intel latest supply update in November, the company apologized for a manufacturing hiccup.
In December, Intel general counsel Steven Rodgers foreshadowed the 5G events the played out throughout the year, as he wrote in an editorial: “Competition will also encourage continued innovation in 5G wireless technology, which will be essential to connected cars, connected health, smarter industry and other applications critical to our future economy. The world benefits from competition in the wireless technology market. We hope that it flourishes.”
It wasn’t all drama, as most of the attention in late 2018 was drawn by its highlight Architecture Day deep-dive event. Intel’s Raja Koduri introduced the company’s six pillars of innovation, from process to software. The company announced multi-year Core and Atom roadmaps and provided the first details of the Sunny Cove microarchitecture and Gen11 graphics inside Ice Lake, and introduced the Xe architecture. Intel had promised the get 10nm systems on shelves by the 2019 holiday season, although some rumors had even pointed to a June launch. The company also announced the 3D die stacking technology Foveros, which it would use in a new product called Lakefield: the company’s first ‘hybrid architecture’ with four Tremont and one Sunny Cove core.
CES 2019: 10nm Innovation Ahead
At CES, Intel introduced the technology (Ice Lake, Lakefield and Foveros) it had announced at Architecture Day to a wider audience. Cooper Lake and Ice Lake would launch for servers in 2020. The 10nm Snow Ridge SoC for 5G base stations was also announced. Its slated 2019 launch hasn’t happened, but we did learn throughout the year that it sports Atom Tremont cores. Intel also announced the Nervana NNP-I for inference to complement the Nervana NNP-T for training. (Intel deep-dived into the chips at HotChips in August.)
In short, Intel’s CES message was one of innovation, with a full lineup of 10nm products planned. Intel also announced Project Athena as a new specification for premium laptops (and the Project Athena Open Labs in May), its 3D Athlete Tracking technology as part of its partnership with the Olympics, shared its plans for 10GbE and Wi-Fi 6, and announced the new F-series of desktop chips with disabled integrated graphics, which received a price cut in October.
In late January, Intel announced record full-year 2018 results. Shortly thereafter, Intel named interim CEO Bob Swan its new CEO, in favor of multiple internal and external candidates that had been circulating in the rumor mill – one of the last ones was Apple chip lead Johny Srouji.
At the end of the year at an annual investor conference, Bob Swan explained his vision with clear words, that the company isn’t just the bigger counterpart of AMD whose Ryzen and Epyc lines might cause it trouble:
"We think about having 30% share in a $230 [silicon] TAM that we think is going to grow to $300B [silicon] TAM over the next 4 years, and frankly, I'm trying to destroy the thinking about having 90% share inside our company because, I think it limits our thinking, I think we miss technology transitions. we miss opportunities because we're, in some ways preoccupied with protecting 90, instead of seeing a much bigger market with much more innovation going on, both inside our four walls, and outside our four walls, so we come to work in the morning with a 30% share, with every expectation over the next several years, that we will play a larger and larger role in our customers success, and that doesn't just mean CPUs.”
Interconnect Day 2019: Compute Express Link and Silicon Photonics
In March, Intel joined the crowed space of cache-coherent interconnect protocols between CPUs and accelerators, competing with the likes of CCIX, OpenCAPI and GenZ: at its Interconnect Day, the company announced the Compute Express Link (CXL) protocol based on PCIe 5.0. Intel also opened it up by forming the CXL Consortium. Products are slated for 2021 across CPUs, GPUs, FPGAs and SmartNICs.
Throughout the year, dozens of companies joined the consortium including AMD, and even Arm announced the transition to CXL from CCIX, which set CXL in the pole position to become the leading coherent interconnect as accelerators become even more important in the next decade.
Silicon photonics also continued gaining momentum, doubling to a two million annual unit run rate. The company also demoed its 400Gbps silicon photonics transceiver and announced plans for an in-between 200Gbps transceiver to launch in 2020.
Data-Centric Innovation Day: Launching Cascade Lake
Intel has been transitioning its strategy to become a data-centric company for several years now, and this notion was reinforced that the company’s major Data-Centric Innovation Day event in April in conjunction with the launch of Cascade Lake-SP and several other product announcements including Agilex 10nm FPGAs, its 800 Series 100Gbps Ethernet adapters and new Xeon-D processors.
Cascade Lake added security fixes, DLBoost and Optane Persistent Memory support to the Purley Xeon Scalable platform. The latter came almost four years after the company’s announcement of 3D XPoint in 2015. Although Micron and Intel separated in the third quarter, both companies remain bullish on the memory.
Exiting 5G Modems
As previous CEO Brian Krzanich had been eager to point out, Intel was the only company with end-to-end 5G capabilities. Intel’s plans were twofold: to develop 5G modem technology and put it in all devices it could think of, and also to invest in the network side to make 5G infrastructure run on Intel servers. Intel had announced its goal for a 40% market share in base stations by 2022 at CES.
On the modem side, however, things quickly fell apart. At MWC, the company’s 5G XMM 8160 modem was still slated for early 2020 launch and widely expected to power that year’s generation of iPhones. In April however, Apple and Qualcomm settled their litigations, which involved a multi-year modem supply agreement among other things (although Apple was rumored to be developing its own modems). In an immediate reaction, Intel exited the 5G smartphone business, as the company saw no more path to profitability anymore with Apple as its only customer.
There had been some discussion as to what was the cause and effect precisely, as Intel reportedly had struggled with its 5G modem effort. Intel’s version is that it exited 5G modems when it learned of the Apple-Qualcomm settlement. As a result of losing Apple as a customer, as Intel explained, its 5G modem big bet did not fulfill all three of its criteria for investment anymore. (Intel’s three criteria are to invest in technology inflections, that impact customers, and also return profits for the company.)
In July, Apple and Intel announced the sale of Intel’s 5G smartphone business to Apple for $1 billion (which recently closed). Intel also sold a portion of its mobile patent portfolio in an auction. Intel retained the option to pursue 5G in other areas, though and in November Intel announced it was partnering with MediaTek to bring 5G to laptops.
In November, Intel filed an amicus brief to the FTC, blaming Qualcomm for its failure in modems.
Investor Meeting 2019: Next-Gen Packaging and Processes
Intel’s next major event happened in May at its first Investor Meeting since early 2017. The company outlined its vision of wanting to lead technology inflections with big bets such as 5G, memory and storage, AI and autonomous vehicles. This involved redefining what Intel inside means: from being a CPU company to being a provider of multiple XPUs such as CPUs, GPUs, FPGAs and AI NPUs.
In a presentation, Murthy outlined his vision of compute evolving from single, monolithic dies to becoming systems much, much larger than the reticle limit, with various XPUs on various process nodes interconnected via advanced packaging. This involved its 2.5D EMIB and 3D Foveros packaging technologies, which Intel added Co-EMIB, MDIO and ODI to in July.
The company also gave a closer look at its process technology plans. Intel admitted that it had been too ambitious with 10nm, combining too many new technologies in one node. The company said it had learned from its mistakes, had reduced its aggressiveness and announced 7nm for production and launch in 2021.
The 7nm process will feature a 2x density increase, EUV and simplified design rules. The company will also continue its practice of developing intra-node optimizations to extract another Moore’s Law of performance and power out of the node.
In October, Intel also mentioned 5nm and its intention to go back to a two-year cadence for the foreseeable future. In December at IEDM, an Intel process roadmap echoed what such a two-year cadence would look like for the full next decade. Intel also detailed some more its research efforts in packaging and process technology, such as a germanium gate-all-around nanoribbon manufactured on top of a silicon FinFET.
Network and Custom Logic Group: Shipping 10nm Agilex
In June, Intel formed the new Network and Custom Logic Group, combining its FPGA group together with its networking group. The new organization aimed to provide customers with the best solutions across the custom logic continuum: Xeon SoCs, FPGAs, eASICs and full custom ASICs. FPGAs are more suitable for fast deployment and innovation, while Intel wants to help brings those to eASICs and ASICs later in the product cycle.
Intel announced that its networking business, reported as part of its data center group in its earnings, was on track to exceed $5 billion in revenue and has become the number one in market share in that segment. On the FPGA side, Intel added two product series to its 14nm Stratix 10 lineup: the Stratix 10 TX early this year became the first FPGA with 58Gbps transceiver technology to support 400Gbps Ethernet, while the Stratix 10 DX became Intel’s cornerstone for accelerating CXL development by supporting Intel’s proprietary UPI interconnect, with the promise of a straightforward migration path to CXL. Intel also announced the huge Stratix 10 GX 10M with 43.4 billion transistors across two large FPGA dies and four transceiver connected via EMIB.
Intel also announced two programmable acceleration cards (PACs) for the data center, although it is doubtful that Intel’s expectation, when it acquired Altera in 2016, of 30% of servers adopting FPGAs by 2020 will be met.
The 10nm Agilex FPGA started shipping in the third quarter, making it Intel’s second shipping 10nm product, in particular noteworthy as FPGAs tend to have a sizeable die size.
Ice Lake CPUs and Graphics
Intel’s second half of the year was characterized by some major announcements in its upcoming growth businesses: progress and plans for its Xe graphics and Mobileye businesses, the launch of its Nervana neural network processor portfolio, and next-gen memory announcements.
https://www.tomshardware.com/features/surface-laptop-3-intel-vs-amd
Ice Lake was finally launched in August. The chip came out favorably against AMD in the latest Surface Laptop. The company also detailed the Atom Tremont architecture. Most notably, Tremont features a dual 3-wide decode unit design and a large out-of-order buffer. On the graphics side, Intel powered on its first-ever DG1 discrete chip.
Memory and Storage Day
…
Mobileye
Mobileye held its inaugural Investor Summit in November, where CEO Amnon Shashua outlined Mobileye’s long-term growth plans in ADAS and autonomous vehicles into 2030. The company uses its profitable ADAS business to fund the development of a self-driving system that it expects to sell for $4000.
The company has shipped its 50 millionth EyeQ SoC at the end of the year, with 17.4 million in 2019 alone. The company also said that 8 out of the 11 L2+ systems in production are based on Mobileye and has 100% design win share on another 13 systems.
Mobileye also plans to go higher up in the value chain, citing a market opportunity of $160 billion, by entering the Mobility-as-a-Service market. It already has plans for Israel, France and China in 2022 followed by the U.S. in 2023. Rather than rushing it, Mobileye is taking a more careful approach to entering the robotaxi market, especially in the US where there are no regulations; 2019 has seen several robotaxi postponements including GM Cruise and Daimler, both based on safety concerns.
BWM’s Level 3-4 iNEXT in 2021 will feature Mobileye’s EyeQ5 and the company has signed a deal with Chinese electric car company NIO to integrate Mobileye’s Level 4 AV Kit, including a variant for Mobileye’s MaaS applications. The Israeli company also announced plans the EyeQ6 to hit the roads in 2023 and for an EyeC portfolio of lidar and radar sensors.
Mobileye remained an independent business after Intel’s acquisition of the company in 2017 – even without financial support – and it is Intel’s smallest reported segment. But if Mobileye’s ambitions become reality a decade from now, third-party and Mobileye-owned robotaxis as well as consumer Level 4-5 AVs with Mobileye silicon, software and sensors will be swerving the planet, guided by its RSS safety model.
AI and oneAPI
Intel launched its long-awaited NNP-I and NNP-T chips for artificial intelligence at the company’s November AI Summit, where it also announced its next-generation Movidius Keem Bay chip. This event preceded the announcement of Ponte Vecchio at Supercomputing 2019: the company’s lead 7nm product that will launch in the fourth quarter of 2021.
At the same event, Raja Koduri also launched the public beta of its oneAPI initiative, which aims to tie together its portfolio of compute capabilities with a unified programming model.
Recent rumors indicate that DG1 might enter the discrete graphics market with a whimper, but Ponte Vecchio might be Intel’s real shot a high-performance GPU. It has already secured an important design win: it will power the Aurora exascale supercomputer.
Quantum and Neuromorphic Computing
Although the company did not announce new quantum or neuromorphic chip this year, it continued research and extended the platforms’ capabilities. Intel scaled its neuromorphic system to 64 of its 60mm2 Loihi chips, codenamed Pohoiki Beach. Intel’s goal is to commercialize neuromorphic (and quantum) computing and it welcomed the first handful of Global 500 companies to its neuromorphic research community.
In quantum computing, Intel is focused on the marathon towards commercial quantum computers with thousands of qubits, although the company applauded Google’s quantum supremacy milestone. To that end, Intel announced the first cryoprober in February called the Cryogenic Wafer Prober, a tool to test qubits quickly at just a few degrees Kelvin. In December, Intel announced a cryogenic control chip for quantum computer called Horse Ridge, which promises to simplify the interconnect issues of the conventional electronics to the cryogenic quantum chip.
Acquisitions
It was a pretty busy year in terms of acquisitions for Intel, with deals across its businesses:
Executive Changes
It was perhaps an even busier year in regards to executive changes, which started of course with the naming of Bob Swan as Intel’s new CEO.
Other changes at the executive management level:
Other notable hires at a lower level:
Departures:
Other Highlights
2020: Looking Ahead
As always, Intel will continue its annual cadence of iterative product launches. Comet Lake will come to the desktop with up to 10 cores, while Intel has Rocket Lake and Tiger Lake for laptops, according to a rumor this year. The same rumor also hinted at Skyhawk Lake (with Tremont cores) in the Atom series. The first Lakefield devices will also enter the shelves, as well as the DG1 chip, 144-layer 3D NAND and 4-layer 3D XPoint.
In the data center, after the over 1.5 years between Skylake and Cascade Lake, Intel has two generations planned in just one year (Intel announced at its investor meeting that it would shorten the timespan between product launches to at most five quarters), besides Snow Ridge: Cooper Lake and Ice Lake. Both have faced delays, as Cooper Lake has slipped from 2019 and Ice Lake from the first to the second half of the year.
Final Words
From a higher level, Intel in 2019 is quite different from the start of this decade, when all its attention shifted to laptops, tablets and smartphones. The company drastically shifted its strategy again around 2015, as it became clear that tablets wouldn’t quite take off and Intel wouldn’t be able to make inroads in smartphones. Instead, Intel focused on its profitable and growing data center, with related areas such as IoT, memory and networking.
With Optane Persistent Memory DIMMs, SmartNICs, and the Nervana portfolio of neural network processors for training and inference, the company entered several brand new markets this year. The company also acquired the Tofino switches and Gaudi and Goya deep learnings chips from Habana. Intel also announced its first graphics cards, DG1 and Ponte Vecchio, and Lakefield with the new Foveros packaging technology.
With those and other product lines such as silicon photonics and the Intel FPGAs, Intel’s mission to diversify its offerings is looking to be successful. Meanwhile, Intel tried fending off competition from AMD by reducing its X-series’ prices by up to 50%.
Also, after a seemingly never ending series of delays, the plagued 10nm process went into volume production and a series of products with Ice Lake have hit the shelves, albeit for laptops only. While Ice Lake is a major upgrade in manufacturing technology and architecture, with Comet Lake (and Cooper Lake for servers) launching in conjunction with Ice Lake, and Rocket Lake on the roadmap, it might not be until some time into 2021 until 10nm overtakes 14nm in volume.
Intel is large chip company with many moving parts, but the its 2029 process roadmap might very well be the standout moment of Intel’s 2019 (although it surfaced through ASML). With a full decade at a two-year cadence, it gives the impression that Moore’s Law is still alive and well, as Intel aims to resurrect its manufacturing execution.
Before the 10nm slips the de facto frontrunner in silicon technology that introduced important and sophisticated process technologies such as FinFET multiple years ahead of others, those 10nm delays have made the battlefield a lot more even. Although the era of gate-all-around transistors in the next decade might lead to more diversity at the transistor level, architecture and even packaging will be at least as important for product leadership in the next decade.
Intel has released its own yearbook here (PDF).
https://newsroom.intel.com/news/2019-yearbook/
Intel 2019 Year in Review
For Intel, 2019 marked an important, albeit turbulent, year. After several years of stagnation on the technology side, the company moved forward with a much-improved architecture on its ‘new’ 10nm process. Other introductions included its Nervana artificial intelligence push, 96-layer 3D NAND and more.
Several storylines also unfolded throughout the year, such as Intel exiting the 5G smartphone modem space, the company’s ongoing struggles with shortages, its plans to return to an aggressive two-year process cadence after 10nm’s three year delay and also plans for a global robotaxi business as part of Mobileye.
Shortages and 5G
These storylines started began in 2018 already. The year started in the wake of the 14nm supply issues, which have continued to persist throughout the year. Intel’s explanation was a combination of unprecedented growth in the data center and a better than expected PC market, the delays of 10nm that resulted in larger 14nm dies to improve performance, and the transition of its chipsets and 4G modems to 14nm and increased modem share.
In response, in December 2018 Intel announced major, multi-dollar factory expansion plans, planned a new $11 billion fab in Israel within the next four years, and progressed on equipping the $9 billion Fab 42 in Arizona for 7nm. When it finally starts production in 2021, it will have been ten years since the fab was announced in 2011. More recently, Intel said it has increased wafer capacity by 25% in 2019 and plans another 25% increase in 2020. In Intel latest supply update in November, the company apologized for a manufacturing hiccup.
In December, Intel general counsel Steven Rodgers foreshadowed the 5G events the played out throughout the year, as he wrote in an editorial: “Competition will also encourage continued innovation in 5G wireless technology, which will be essential to connected cars, connected health, smarter industry and other applications critical to our future economy. The world benefits from competition in the wireless technology market. We hope that it flourishes.”
It wasn’t all drama, as most of the attention in late 2018 was drawn by its highlight Architecture Day deep-dive event. Intel’s Raja Koduri introduced the company’s six pillars of innovation, from process to software. The company announced multi-year Core and Atom roadmaps and provided the first details of the Sunny Cove microarchitecture and Gen11 graphics inside Ice Lake, and introduced the Xe architecture. Intel had promised the get 10nm systems on shelves by the 2019 holiday season, although some rumors had even pointed to a June launch. The company also announced the 3D die stacking technology Foveros, which it would use in a new product called Lakefield: the company’s first ‘hybrid architecture’ with four Tremont and one Sunny Cove core.
CES 2019: 10nm Innovation Ahead
At CES, Intel introduced the technology (Ice Lake, Lakefield and Foveros) it had announced at Architecture Day to a wider audience. Cooper Lake and Ice Lake would launch for servers in 2020. The 10nm Snow Ridge SoC for 5G base stations was also announced. Its slated 2019 launch hasn’t happened, but we did learn throughout the year that it sports Atom Tremont cores. Intel also announced the Nervana NNP-I for inference to complement the Nervana NNP-T for training. (Intel deep-dived into the chips at HotChips in August.)
In short, Intel’s CES message was one of innovation, with a full lineup of 10nm products planned. Intel also announced Project Athena as a new specification for premium laptops (and the Project Athena Open Labs in May), its 3D Athlete Tracking technology as part of its partnership with the Olympics, shared its plans for 10GbE and Wi-Fi 6, and announced the new F-series of desktop chips with disabled integrated graphics, which received a price cut in October.
In late January, Intel announced record full-year 2018 results. Shortly thereafter, Intel named interim CEO Bob Swan its new CEO, in favor of multiple internal and external candidates that had been circulating in the rumor mill – one of the last ones was Apple chip lead Johny Srouji.
At the end of the year at an annual investor conference, Bob Swan explained his vision with clear words, that the company isn’t just the bigger counterpart of AMD whose Ryzen and Epyc lines might cause it trouble:
"We think about having 30% share in a $230 [silicon] TAM that we think is going to grow to $300B [silicon] TAM over the next 4 years, and frankly, I'm trying to destroy the thinking about having 90% share inside our company because, I think it limits our thinking, I think we miss technology transitions. we miss opportunities because we're, in some ways preoccupied with protecting 90, instead of seeing a much bigger market with much more innovation going on, both inside our four walls, and outside our four walls, so we come to work in the morning with a 30% share, with every expectation over the next several years, that we will play a larger and larger role in our customers success, and that doesn't just mean CPUs.”
Interconnect Day 2019: Compute Express Link and Silicon Photonics
In March, Intel joined the crowed space of cache-coherent interconnect protocols between CPUs and accelerators, competing with the likes of CCIX, OpenCAPI and GenZ: at its Interconnect Day, the company announced the Compute Express Link (CXL) protocol based on PCIe 5.0. Intel also opened it up by forming the CXL Consortium. Products are slated for 2021 across CPUs, GPUs, FPGAs and SmartNICs.
Throughout the year, dozens of companies joined the consortium including AMD, and even Arm announced the transition to CXL from CCIX, which set CXL in the pole position to become the leading coherent interconnect as accelerators become even more important in the next decade.
Silicon photonics also continued gaining momentum, doubling to a two million annual unit run rate. The company also demoed its 400Gbps silicon photonics transceiver and announced plans for an in-between 200Gbps transceiver to launch in 2020.
Data-Centric Innovation Day: Launching Cascade Lake
Intel has been transitioning its strategy to become a data-centric company for several years now, and this notion was reinforced that the company’s major Data-Centric Innovation Day event in April in conjunction with the launch of Cascade Lake-SP and several other product announcements including Agilex 10nm FPGAs, its 800 Series 100Gbps Ethernet adapters and new Xeon-D processors.
Cascade Lake added security fixes, DLBoost and Optane Persistent Memory support to the Purley Xeon Scalable platform. The latter came almost four years after the company’s announcement of 3D XPoint in 2015. Although Micron and Intel separated in the third quarter, both companies remain bullish on the memory.
Exiting 5G Modems
As previous CEO Brian Krzanich had been eager to point out, Intel was the only company with end-to-end 5G capabilities. Intel’s plans were twofold: to develop 5G modem technology and put it in all devices it could think of, and also to invest in the network side to make 5G infrastructure run on Intel servers. Intel had announced its goal for a 40% market share in base stations by 2022 at CES.
On the modem side, however, things quickly fell apart. At MWC, the company’s 5G XMM 8160 modem was still slated for early 2020 launch and widely expected to power that year’s generation of iPhones. In April however, Apple and Qualcomm settled their litigations, which involved a multi-year modem supply agreement among other things (although Apple was rumored to be developing its own modems). In an immediate reaction, Intel exited the 5G smartphone business, as the company saw no more path to profitability anymore with Apple as its only customer.
There had been some discussion as to what was the cause and effect precisely, as Intel reportedly had struggled with its 5G modem effort. Intel’s version is that it exited 5G modems when it learned of the Apple-Qualcomm settlement. As a result of losing Apple as a customer, as Intel explained, its 5G modem big bet did not fulfill all three of its criteria for investment anymore. (Intel’s three criteria are to invest in technology inflections, that impact customers, and also return profits for the company.)
In July, Apple and Intel announced the sale of Intel’s 5G smartphone business to Apple for $1 billion (which recently closed). Intel also sold a portion of its mobile patent portfolio in an auction. Intel retained the option to pursue 5G in other areas, though and in November Intel announced it was partnering with MediaTek to bring 5G to laptops.
In November, Intel filed an amicus brief to the FTC, blaming Qualcomm for its failure in modems.
Investor Meeting 2019: Next-Gen Packaging and Processes
Intel’s next major event happened in May at its first Investor Meeting since early 2017. The company outlined its vision of wanting to lead technology inflections with big bets such as 5G, memory and storage, AI and autonomous vehicles. This involved redefining what Intel inside means: from being a CPU company to being a provider of multiple XPUs such as CPUs, GPUs, FPGAs and AI NPUs.
In a presentation, Murthy outlined his vision of compute evolving from single, monolithic dies to becoming systems much, much larger than the reticle limit, with various XPUs on various process nodes interconnected via advanced packaging. This involved its 2.5D EMIB and 3D Foveros packaging technologies, which Intel added Co-EMIB, MDIO and ODI to in July.
The company also gave a closer look at its process technology plans. Intel admitted that it had been too ambitious with 10nm, combining too many new technologies in one node. The company said it had learned from its mistakes, had reduced its aggressiveness and announced 7nm for production and launch in 2021.
The 7nm process will feature a 2x density increase, EUV and simplified design rules. The company will also continue its practice of developing intra-node optimizations to extract another Moore’s Law of performance and power out of the node.
In October, Intel also mentioned 5nm and its intention to go back to a two-year cadence for the foreseeable future. In December at IEDM, an Intel process roadmap echoed what such a two-year cadence would look like for the full next decade. Intel also detailed some more its research efforts in packaging and process technology, such as a germanium gate-all-around nanoribbon manufactured on top of a silicon FinFET.
Network and Custom Logic Group: Shipping 10nm Agilex
In June, Intel formed the new Network and Custom Logic Group, combining its FPGA group together with its networking group. The new organization aimed to provide customers with the best solutions across the custom logic continuum: Xeon SoCs, FPGAs, eASICs and full custom ASICs. FPGAs are more suitable for fast deployment and innovation, while Intel wants to help brings those to eASICs and ASICs later in the product cycle.
Intel announced that its networking business, reported as part of its data center group in its earnings, was on track to exceed $5 billion in revenue and has become the number one in market share in that segment. On the FPGA side, Intel added two product series to its 14nm Stratix 10 lineup: the Stratix 10 TX early this year became the first FPGA with 58Gbps transceiver technology to support 400Gbps Ethernet, while the Stratix 10 DX became Intel’s cornerstone for accelerating CXL development by supporting Intel’s proprietary UPI interconnect, with the promise of a straightforward migration path to CXL. Intel also announced the huge Stratix 10 GX 10M with 43.4 billion transistors across two large FPGA dies and four transceiver connected via EMIB.
Intel also announced two programmable acceleration cards (PACs) for the data center, although it is doubtful that Intel’s expectation, when it acquired Altera in 2016, of 30% of servers adopting FPGAs by 2020 will be met.
The 10nm Agilex FPGA started shipping in the third quarter, making it Intel’s second shipping 10nm product, in particular noteworthy as FPGAs tend to have a sizeable die size.
Ice Lake CPUs and Graphics
Intel’s second half of the year was characterized by some major announcements in its upcoming growth businesses: progress and plans for its Xe graphics and Mobileye businesses, the launch of its Nervana neural network processor portfolio, and next-gen memory announcements.
https://www.tomshardware.com/features/surface-laptop-3-intel-vs-amd
Ice Lake was finally launched in August. The chip came out favorably against AMD in the latest Surface Laptop. The company also detailed the Atom Tremont architecture. Most notably, Tremont features a dual 3-wide decode unit design and a large out-of-order buffer. On the graphics side, Intel powered on its first-ever DG1 discrete chip.
Memory and Storage Day
…
Mobileye
Mobileye held its inaugural Investor Summit in November, where CEO Amnon Shashua outlined Mobileye’s long-term growth plans in ADAS and autonomous vehicles into 2030. The company uses its profitable ADAS business to fund the development of a self-driving system that it expects to sell for $4000.
The company has shipped its 50 millionth EyeQ SoC at the end of the year, with 17.4 million in 2019 alone. The company also said that 8 out of the 11 L2+ systems in production are based on Mobileye and has 100% design win share on another 13 systems.
Mobileye also plans to go higher up in the value chain, citing a market opportunity of $160 billion, by entering the Mobility-as-a-Service market. It already has plans for Israel, France and China in 2022 followed by the U.S. in 2023. Rather than rushing it, Mobileye is taking a more careful approach to entering the robotaxi market, especially in the US where there are no regulations; 2019 has seen several robotaxi postponements including GM Cruise and Daimler, both based on safety concerns.
BWM’s Level 3-4 iNEXT in 2021 will feature Mobileye’s EyeQ5 and the company has signed a deal with Chinese electric car company NIO to integrate Mobileye’s Level 4 AV Kit, including a variant for Mobileye’s MaaS applications. The Israeli company also announced plans the EyeQ6 to hit the roads in 2023 and for an EyeC portfolio of lidar and radar sensors.
Mobileye remained an independent business after Intel’s acquisition of the company in 2017 – even without financial support – and it is Intel’s smallest reported segment. But if Mobileye’s ambitions become reality a decade from now, third-party and Mobileye-owned robotaxis as well as consumer Level 4-5 AVs with Mobileye silicon, software and sensors will be swerving the planet, guided by its RSS safety model.
AI and oneAPI
Intel launched its long-awaited NNP-I and NNP-T chips for artificial intelligence at the company’s November AI Summit, where it also announced its next-generation Movidius Keem Bay chip. This event preceded the announcement of Ponte Vecchio at Supercomputing 2019: the company’s lead 7nm product that will launch in the fourth quarter of 2021.
At the same event, Raja Koduri also launched the public beta of its oneAPI initiative, which aims to tie together its portfolio of compute capabilities with a unified programming model.
Recent rumors indicate that DG1 might enter the discrete graphics market with a whimper, but Ponte Vecchio might be Intel’s real shot a high-performance GPU. It has already secured an important design win: it will power the Aurora exascale supercomputer.
Quantum and Neuromorphic Computing
Although the company did not announce new quantum or neuromorphic chip this year, it continued research and extended the platforms’ capabilities. Intel scaled its neuromorphic system to 64 of its 60mm2 Loihi chips, codenamed Pohoiki Beach. Intel’s goal is to commercialize neuromorphic (and quantum) computing and it welcomed the first handful of Global 500 companies to its neuromorphic research community.
In quantum computing, Intel is focused on the marathon towards commercial quantum computers with thousands of qubits, although the company applauded Google’s quantum supremacy milestone. To that end, Intel announced the first cryoprober in February called the Cryogenic Wafer Prober, a tool to test qubits quickly at just a few degrees Kelvin. In December, Intel announced a cryogenic control chip for quantum computer called Horse Ridge, which promises to simplify the interconnect issues of the conventional electronics to the cryogenic quantum chip.
Acquisitions
It was a pretty busy year in terms of acquisitions for Intel, with deals across its businesses:
- Omnitek for vision applications on FPGAs
- Barefoot Networks silicon switch startup
- Smart Edge for 5G edge computing
- Indian SoC designer Ineda Systems, by outsiders described as an acquihire, an acquisition for the workforce’ skills in graphics; Intel confirmed it would help the company’s discrete GPU business
- AI inference and training chip startup Habana Labs for $2 billion, which notably overlaps with its current Nervana business
Executive Changes
It was perhaps an even busier year in regards to executive changes, which started of course with the naming of Bob Swan as Intel’s new CEO.
Other changes at the executive management level:
- George Davis became Intel’s new CFO, succeeding Bob Swan and joining from Qualcomm
- Gregory Bryant was promoted from corporate to executive vice president of Intel Client Computing Group
- Sandra Rivera, previously responsible for the company’s Network Platforms Group, became Chief People Officer (as a consequence of the group’s merging with the FPGA group)
- Craig Barratt from Barefoot Networks joined the executive leadership
- Claire Dixon from VMWare became Chief Communications Officer
- Karen Walker from Cisco became Intel’s new Chief Marketing Officer
Other notable hires at a lower level:
- Gary Patton, previously GlobalFoundries’ CTO, joined Intel in December to lead design enablement of process technologies
- Johann Jungwirth was appointed as vice president of Mobility-as-a-Service (MaaS) at Mobileye, previously executive vice president for mobility services at Volkswagen
- Masooma Bhaiwala, previously vice president of semi-custom SoCs at AMD, became Intel’s vice president of discrete GPU SoCs
- Devon Nekechuk, who previously served as Senior Manager of Product Management at AMD, became Intel’s Graphics Product Director
- Tom Petersen, previously Nvidia’s director of technical marketing, become an Intel Fellow working with the Game Experience Team
Departures:
- Chris Hook and Heather Lennon, both previously responsible for AMD marketing, left Intel in November
- Cormac Conroy, corporate vice president of the Communication and Devices Group, resigned, possibly as its 5G efforts dwindled
- Praveen Vishakantaiah, VP of Client Architecture and Innovation, left after a 26-year career at Intel
- Imad Sousou, corporate vice president of System Software Products, left the company after serving over 25 years for Intel
- Rajeeb Hazra, corporate vice president of the Data Center Group’s Enterprise and Government Group, left after 24 years at Intel
Other Highlights
- Also in the first half of 2019, Intel contributed the Thunderbolt 3 protocol to the USB Promoter Group, which is being used for USB4
- Intel and Netflix announced the SVT-AV1 codec
- Intel launched the tenth-generation Ice Lake-U and Comet Lake-U in August. For the data center, Cooper Lake would consists of up to 56 cores
- Intel added two RealSense cameras to its portfolio: the RealSense Tracking Camera T265, announced in January, features a Movidius Myriad 2 VPU and is aimed at tracking the location of autonomous devices in locations without GPS service; the RealSense lidar camera L515 was announced this month and marked the company’s inroads in lidar technology products
- Omni-Path developed stopped
2020: Looking Ahead
As always, Intel will continue its annual cadence of iterative product launches. Comet Lake will come to the desktop with up to 10 cores, while Intel has Rocket Lake and Tiger Lake for laptops, according to a rumor this year. The same rumor also hinted at Skyhawk Lake (with Tremont cores) in the Atom series. The first Lakefield devices will also enter the shelves, as well as the DG1 chip, 144-layer 3D NAND and 4-layer 3D XPoint.
In the data center, after the over 1.5 years between Skylake and Cascade Lake, Intel has two generations planned in just one year (Intel announced at its investor meeting that it would shorten the timespan between product launches to at most five quarters), besides Snow Ridge: Cooper Lake and Ice Lake. Both have faced delays, as Cooper Lake has slipped from 2019 and Ice Lake from the first to the second half of the year.
Final Words
From a higher level, Intel in 2019 is quite different from the start of this decade, when all its attention shifted to laptops, tablets and smartphones. The company drastically shifted its strategy again around 2015, as it became clear that tablets wouldn’t quite take off and Intel wouldn’t be able to make inroads in smartphones. Instead, Intel focused on its profitable and growing data center, with related areas such as IoT, memory and networking.
With Optane Persistent Memory DIMMs, SmartNICs, and the Nervana portfolio of neural network processors for training and inference, the company entered several brand new markets this year. The company also acquired the Tofino switches and Gaudi and Goya deep learnings chips from Habana. Intel also announced its first graphics cards, DG1 and Ponte Vecchio, and Lakefield with the new Foveros packaging technology.
With those and other product lines such as silicon photonics and the Intel FPGAs, Intel’s mission to diversify its offerings is looking to be successful. Meanwhile, Intel tried fending off competition from AMD by reducing its X-series’ prices by up to 50%.
Also, after a seemingly never ending series of delays, the plagued 10nm process went into volume production and a series of products with Ice Lake have hit the shelves, albeit for laptops only. While Ice Lake is a major upgrade in manufacturing technology and architecture, with Comet Lake (and Cooper Lake for servers) launching in conjunction with Ice Lake, and Rocket Lake on the roadmap, it might not be until some time into 2021 until 10nm overtakes 14nm in volume.
Intel is large chip company with many moving parts, but the its 2029 process roadmap might very well be the standout moment of Intel’s 2019 (although it surfaced through ASML). With a full decade at a two-year cadence, it gives the impression that Moore’s Law is still alive and well, as Intel aims to resurrect its manufacturing execution.
Before the 10nm slips the de facto frontrunner in silicon technology that introduced important and sophisticated process technologies such as FinFET multiple years ahead of others, those 10nm delays have made the battlefield a lot more even. Although the era of gate-all-around transistors in the next decade might lead to more diversity at the transistor level, architecture and even packaging will be at least as important for product leadership in the next decade.
Intel has released its own yearbook here (PDF).
https://newsroom.intel.com/news/2019-yearbook/
Facebook
Twitter