SK Hynix’s 1y DRAM Node: Examining the Company’s Second-Generation 10 nm-Class Process Technology

The core topic is SK Hynix’s 1y (or 1Ynm) DRAM process node—the second generation of its 10 nm-class manufacturing technology—and what it represented in the evolution of DRAM scaling, performance, productivity, and product introductions around 2018–2019. Ambiguities in public reporting include exact half-pitch or design-rule numbers (industry “10 nm-class” labels are marketing ranges rather than precise measured dimensions) and limited open technical cross-sections compared with later EUV-era nodes; the discussion therefore draws on company announcements, contemporaneous industry analyses, and the established generational sequence.

Context Within the 10 nm-Class Roadmap

DRAM makers moved from 20 nm-class nodes into the broadly labeled “10 nm class” in the mid-to-late 2010s. SK Hynix’s sequence within this class follows the common industry convention of successive letter designations:

• 1x — first generation (~18 nm range)
• 1y — second generation (~17 nm range)
• 1z — third generation (~15–16 nm range)
• 1a — fourth generation (~13–14 nm range; first broad EUV adoption at SK Hynix)
• 1b — fifth generation (~12–13 nm)
• 1c — sixth generation (~11–12 nm; world-first development announced by SK Hynix in 2024)

The 1y node therefore sits as a critical mid-class bridge: it delivered meaningful density and efficiency gains after the initial 1x transition while still relying primarily on multi-patterning techniques rather than the extreme-ultraviolet (EUV) lithography that became more prominent from 1a onward.

SK Hynix publicly announced development of its 1Ynm 8 Gb DDR4 DRAM in November 2018 and, in the same month, the industry’s first JEDEC-compliant 16 Gb DDR5 DRAM also fabricated on the 1Ynm process. Shipping of the DDR4 product was targeted for the first quarter of 2019, with broader expansion into server, PC, and later mobile applications.

Key Technical and Product Claims for the 1y Node

Compared with the preceding 1Xnm generation, SK Hynix reported the following improvements for the 1Ynm 8 Gb DDR4:

• Approximately 20 % higher productivity (more usable dies per wafer), reflecting tighter design rules and process optimization.
• More than 15 % reduction in power consumption.
• Support for data rates up to 3,200 Mbps—the highest then available in the DDR4 interface.

Circuit-level innovations included a 4-phase clocking scheme that effectively doubles the clock signal for improved transfer speed and stability, plus proprietary “Sense Amp. Control” technology. The latter refined the sense-amplifier transistor structure and incorporated low-power supply circuitry to cut both power draw and data-error susceptibility—challenges that intensify as feature sizes shrink.

The concurrent 16 Gb DDR5 device on the same 1Ynm process lowered operating voltage from 1.2 V (typical of DDR4) to 1.1 V, yielding roughly 30 % lower power than DDR4 while delivering ~5,200 Mbps transfer rates—about 60 % faster than the 3,200 Mbps DDR4 baseline of the era. This positioned SK Hynix early in the DDR5 transition, with mass production planned for 2020 when market demand was expected to ramp.

Manufacturing and Scaling Considerations

At the 1y generation, manufacturers continued to rely heavily on immersion multi-patterning (including techniques related to quadruple patterning used earlier in the 10 nm class) rather than full EUV for critical layers. SK Hynix later noted partial EUV use on 1ynm products as a learning step before committing to broader EUV adoption for 1a-node mass production in 2021. Capacitor and transistor structures continued to evolve—industry analyses of contemporaneous nodes show ongoing refinement of cylindrical or quasi-cylindrical capacitors, high-k dielectrics, and recessed-channel or buried-wordline transistors to maintain adequate cell capacitance and reliability as lateral dimensions shrank.

Die-size and bit-density gains relative to 1x were meaningful but already showing the diminishing returns that later nodes would confront more acutely. Productivity improvements of ~20 % were competitive yet lagged some claims from peers in certain product segments, illustrating that “same-named” nodes from different suppliers can differ in exact half-pitch, cell size, and net die count.

Competitive Landscape and Industry Timing

Samsung had begun mass production of its own 1y-class DRAM earlier (late 2017), while Micron and SK Hynix both publicized second-generation 10 nm-class devices around the same November 2018 window. Differences in measured cell size, bit density, and wafer productivity among the three major suppliers were noted by independent analysts, underscoring that process-node labels are approximate and that cost and performance leadership depended on a combination of lithography strategy, materials engineering, and design efficiency.

By 2019 SK Hynix had already moved on to announce 1Znm 16 Gb DDR4, claiming further productivity (~27 %) and power (~40 % relative to prior 8 Gb modules) gains without requiring expensive EUV at that stage. The 1y node therefore functioned as a high-volume workhorse that supported the company’s early DDR5 leadership claim and provided a stable platform while next-generation process development continued.

Broader Implications for DRAM Evolution

The 1y generation illustrates several recurring themes in DRAM scaling:

• Productivity versus performance trade-offs — each successive shrink must deliver enough extra dies per wafer to offset rising process complexity and capital intensity.
• Power and reliability engineering — as cells shrink, leakage, sensing margins, and data integrity become first-order concerns addressed through circuit techniques (sense-amp control, multi-phase clocking) as much as pure lithography.
• Product-roadmap leverage — applying a new node simultaneously to mainstream DDR4 and next-generation DDR5 maximized return on process investment and gave customers a smoother migration path.
• Transition toward EUV — 1y served as a partial-EUV proving ground, preparing manufacturing and design teams for the denser 1a and later nodes that underpinned subsequent HBM, LPDDR5X/T, and high-capacity server DRAM.

In retrospect, the 1y node helped SK Hynix maintain competitive density and efficiency while the industry collectively pushed deeper into the single-digit nanometer regime. Subsequent generations (especially 1a onward with fuller EUV utilization, and the later 1c node) continued the trajectory toward higher bit density, lower energy per bit, and specialized products for AI and high-performance computing. The same fundamental challenges—cell capacitance retention, patterning fidelity, and cost per bit—remain central today as the industry explores 4F² vertical-gate architectures and longer-term 3D DRAM concepts beyond conventional 6F² planar scaling.

Overall, SK Hynix’s 1y DRAM process stands as a well-documented example of incremental yet commercially significant process refinement: a second-generation 10 nm-class technology that delivered measurable gains in productivity, power, and speed, enabled early JEDEC-compliant DDR5 silicon, and formed a practical bridge between the first 10 nm-class devices and the EUV-intensive nodes that followed.

---

1) Key Technical and Product Claims

8 Gb DDR4 DRAM on 1Ynm

SK Hynix positioned the 1Ynm 8 Gb DDR4 as a direct successor to its prior 1Xnm generation, with the following quantified improvements:

• Productivity: Approximately 20 % higher number of usable dies per wafer compared with the previous-generation 1Xnm DRAM. This reflected tighter design rules and overall process optimization within the second-generation 10 nm-class node.
• Power consumption: More than 15 % reduction relative to the 1Xnm predecessor.
• Data transfer rate: Support for up to 3,200 Mbps, described at the time as the fastest processing speed available in the DDR4 interface.

Circuit and Design Innovations Cited

• 4-Phase Clocking scheme: Doubles the effective clock signal to improve both data-transfer speed and signal stability.
• Proprietary Sense Amp. Control technology: Enhanced sense-amplifier performance by refining the transistor structure (to reduce data-error susceptibility that increases with feature-size shrink) and by adding a dedicated low-power supply circuit that limits unnecessary power draw.

The company stated that the device offered “optimum performance and density” for customers and planned initial shipments in the first quarter of 2019, beginning with server and PC segments before later expansion into mobile applications.

16 Gb DDR5 DRAM on the Same 1Ynm Process

Announced only days later, the 16 Gb DDR5 device applied the identical 1Ynm process and was presented as the industry’s first JEDEC-compliant DDR5 DRAM. Key claims included:

• Operating voltage: Reduced from the 1.2 V typical of DDR4 to 1.1 V.
• Power consumption: Approximately 30 % lower than prior-generation DDR4 DRAM.
• Data transfer rate: 5,200 Mbps—about 60 % faster than the 3,200 Mbps baseline of contemporary DDR4—enabling processing of 41.6 GB of data per second (equivalent to roughly 11 full-HD video files of 3.7 GB each).
• Bank architecture: Doubled the number of memory banks from 16 to 32 in accordance with JEDEC DDR5 standards, supporting higher concurrent access.

SK Hynix supplied RDIMM and UDIMM samples to a major chipset maker for server and PC platform validation. Mass production was projected for 2020, timed to expected market demand growth for DDR5.

Summary of Core Value Propositions

Aspect	8 Gb DDR4 (1Ynm)	16 Gb DDR5 (1Ynm)
Productivity gain	~20 % vs. 1Xnm	Same process base
Power reduction	>15 % vs. 1Xnm	~30 % vs. DDR4
Peak data rate	Up to 3,200 Mbps	5,200 Mbps (~60 % faster than DDR4)
Voltage	Standard DDR4	1.1 V (from 1.2 V)
Key circuit features	4-phase clocking; Sense Amp. Control	JEDEC-compliant bank doubling; process reuse
Target markets	Server/PC first, then mobile	Server/PC platforms; early DDR5 leadership

These claims collectively framed the 1y node as delivering measurable manufacturing efficiency, lower energy per bit, and competitive interface speeds while enabling an early, standards-compliant entry into DDR5. Independent analyses later confirmed that first-generation commercial DDR5 devices from multiple suppliers, including SK Hynix, often leveraged 1y-class or similar mid-10 nm processes before denser EUV-based nodes became dominant. The public technical narrative remained focused on the productivity, power, and speed metrics outlined above rather than granular process metrics such as exact word-line or bit-line pitches.

---

2) Manufacturing and Scaling Considerations

The core focus is the manufacturing processes, lithography approaches, cell and capacitor integration challenges, and scaling dynamics associated with SK Hynix’s second-generation 10 nm-class (1y / 1Ynm) DRAM technology. This node, developed and announced around 2018, occupied a transitional position in the industry’s progression through the 10 nm class: it delivered measurable density and efficiency gains after the initial 1x generation while still relying predominantly on established deep-ultraviolet (DUV) multi-patterning rather than the broader extreme-ultraviolet (EUV) adoption that began in earnest at the later 1a node. Public technical details remain limited compared with more recent nodes, but available company statements and independent analyses outline the key process characteristics and physical constraints.

Lithography and Patterning Strategy

At the 1y generation, SK Hynix and its peers continued to extend 193 nm immersion lithography through multiple patterning techniques (self-aligned double patterning or SADP, litho-etch-litho-etch or LELE, and related variants) to define critical layers such as active areas, word lines, bit lines, and storage-node contacts. Full EUV was not yet the primary tool for high-volume production on this node. SK Hynix later described partial EUV adoption on certain 1ynm layers as a learning exercise that helped de-risk broader EUV integration for the subsequent 1a node (fourth-generation 10 nm-class), which entered mass production in 2021.

This multi-patterning approach increased process complexity and mask count relative to earlier nodes, yet it remained cost-effective compared with the capital intensity of early EUV tools. Industry analyses of the broader 10 nm-class era note that DUV multi-patterning could still support scaling through much of the class, though the number of patterning steps rose sharply at later generations (particularly from 1b onward), creating pressure to introduce EUV for critical layers to reduce cycle time and improve pattern fidelity.

Cell Architecture and Capacitor Integration

DRAM cells at this scale retained the conventional 6F² layout with buried word-line (or similar recessed-channel) access transistors and capacitor-over-bit-line structures. Maintaining adequate cell capacitance (typically targeted above roughly 7–10 fF per cell in earlier analyses of the era) while shrinking lateral dimensions required continuous refinement of the storage capacitor.

Independent process analyses indicate that SK Hynix transitioned capacitor integration from purely cylindrical structures toward quasi-cylindrical designs around the D1y and D1z generations. High-k dielectric stacks were thinned and optimized (often based on zirconium or hafnium oxides with aluminum doping or similar engineering), and electrode materials evolved to support higher aspect-ratio structures without collapse. These changes helped preserve sensing margins as the physical volume available for the capacitor decreased. Concurrently, sense-amplifier transistor structures were adjusted (for example, recessed-channel variants) to improve reliability under tighter design rules.

Bit-line and word-line pitches continued to scale, though the shrink factor (ratio of successive feature sizes) was already increasing generation by generation, signaling the onset of diminishing returns that would become more pronounced at 1z and beyond.

Key Scaling Challenges at the 1y Node

Several interrelated physical and process limitations characterized manufacturing at this stage:

• Capacitance retention versus lateral shrink — Reducing cell area lowers the physical volume available for the storage capacitor. Maintaining sufficient charge storage demanded taller or more complex high-aspect-ratio structures and thinner, higher-permittivity dielectrics, both of which increased the risk of mechanical instability or leakage.
• Patterning fidelity and overlay control — Multi-patterning introduces additional opportunities for edge-placement error and critical-dimension variation. Tighter pitches amplified the impact of any misalignment on yield and performance.
• Transistor reliability and leakage — Smaller channel dimensions and higher electric fields raised concerns about short-channel effects, data retention, and disturbance phenomena. Circuit-level mitigations (such as the Sense Amp. Control techniques claimed for the 1Ynm products) complemented process changes.
• Productivity versus complexity trade-off — The reported ~20 % productivity gain for the 1Ynm 8 Gb DDR4 relative to 1Xnm reflected successful net-die improvement, yet each successive shrink required more process steps and tighter process control, raising manufacturing cost per wafer even as bits per wafer increased.
• Transition readiness for EUV — Partial EUV use on 1ynm provided early manufacturing experience with the new light source, mask infrastructure, and resist chemistry, easing the later ramp of 1a production where EUV became more extensively applied.

Positioning Within the Broader Scaling Trajectory

The 1y node exemplified the industry’s strategy of extracting additional value from multi-patterning before the full economic and technical case for EUV matured. Subsequent nodes (1z without heavy EUV, then 1a with selective EUV, and later 1b/1c with expanding EUV layer counts) progressively increased EUV utilization to manage the rising complexity of patterning and to improve yield at tighter pitches. By the mid-2020s, analyses indicated that conventional 6F² planar scaling was approaching practical limits near the low-10 nm regime, prompting exploration of 4F² vertical-gate architectures and longer-term 3D DRAM concepts.

In summary, manufacturing the 1y node required careful optimization of established DUV multi-patterning flows, continued evolution of high-aspect-ratio capacitor processes, and incremental transistor and circuit refinements to sustain performance and reliability. These efforts delivered the productivity, power, and speed gains claimed for the contemporaneous DDR4 and early DDR5 products while serving as a practical bridge toward the more EUV-intensive generations that followed. The same fundamental constraints—capacitance scaling, patterning complexity, and cost per bit—continue to shape DRAM process development today.

---

3) Competitive Landscape and Industry Timing

Relative Timing of 1y Node Introductions

Samsung established an early lead on the 1y node. The company began mass production of its second-generation 10 nm-class 8 Gb DDR4 DRAM in November 2017 and followed with LPDDR4X and later LPDDR5 products on the same process family. By mid-2019 Samsung was already shipping high-capacity mobile packages (including 12 GB LPDDR4X) built on 1y-nm dies.

SK Hynix publicly announced development of its 1Ynm 8 Gb DDR4 and the industry’s first JEDEC-compliant 16 Gb DDR5 in November 2018, with initial DDR4 shipments planned for the first quarter of 2019. The company positioned the node for both server/PC DDR4 and as the foundation for its early DDR5 ramp (commercial volume beginning in 2020).

Micron targeted initial production of its 1Y-nm DRAM in the second half of 2018. In November 2018 the company announced mass production of the industry’s first monolithic 12 Gb LPDDR4x on the 1Y-nm process, emphasizing capacity and power advantages for mobile applications. Micron’s 1X-nm bit crossover was still progressing through 2018, so the 1Y transition occurred while 1X remained a significant portion of output.

In short, Samsung led by roughly a year in volume production of 1y-class DRAM, while SK Hynix and Micron moved into the node in late 2018 with overlapping but differentiated product focuses.

Productivity and Cost Competitiveness Differences

Even when labeled the same “1y” generation, the three suppliers did not achieve identical die sizes, bit densities, or net dies per wafer. Contemporary industry reporting noted that Samsung’s 1y productivity gain versus its own 1x was claimed at approximately 30 percent, giving it the highest net-die count among the three. SK Hynix reported a 20 percent productivity increase for its 1Ynm 8 Gb DDR4 versus 1Xnm. Micron’s relative position was generally viewed as trailing the two Korean suppliers in net dies at the early 1y stage, consistent with historical patterns in which process-node labels mask differences in actual design rules, cell size, and yield maturity.

These differences translated directly into cost-per-bit competitiveness. Higher net dies lowered manufacturing cost and improved gross margins once yields stabilized. Samsung’s earlier ramp therefore conferred a temporary cost and volume advantage in both DDR4 and mobile segments, while SK Hynix leveraged its 1y process to secure an early standards-compliant DDR5 position.

Product-Segment Strategies and Market Positioning

• Samsung prioritized mobile LPDDR4X and LPDDR5 packages, using the density of 1y to enable higher-capacity modules (8 GB and 12 GB) that supported flagship smartphone features such as multi-camera systems and larger displays. The company also maintained a strong presence in server and graphics DRAM.
• SK Hynix used the 1y node to bridge mainstream DDR4 (emphasizing 3,200 Mbps speed and power reduction) and to launch the first JEDEC-compliant DDR5, giving it a visible leadership claim in the next-generation server and PC memory transition. Expansion into mobile followed.
• Micron focused initial 1Y volume on high-capacity monolithic mobile LPDDR4x (12 Gb), highlighting power savings of up to 10 percent at 4,266 Mbps data rates. The company continued parallel ramps of 1X for broader DDR4 coverage while preparing subsequent nodes.

This segmentation meant that “first-to-market” claims were often product-specific rather than pure process-node races. SK Hynix’s early DDR5 announcement, for example, carried strategic weight even if its overall 1y wafer starts lagged Samsung’s.

Broader Industry Context and Subsequent Transitions

The 1y generation occurred against a backdrop of strong DRAM demand in 2017–2018, driven by server expansion, smartphone content growth, and the early stages of AI/data-center workloads. All three suppliers were simultaneously managing the residual 20 nm-class and 1x output while investing in the more complex multi-patterning required for 1y and the forthcoming 1z node.

By 2019–2020 the competitive dynamic had already shifted toward 1z (third-generation 10 nm-class). Samsung began 1z mass production in 2019, SK Hynix announced 1Znm 16 Gb DDR4 development in late 2019, and Micron advanced its own 1Z roadmap. The 1y node therefore functioned as a relatively short-lived high-volume platform—especially for SK Hynix—before the industry moved on to tighter design rules and the gradual introduction of EUV lithography at the 1a node.

Independent reverse-engineering work later confirmed that early commercial DDR5 devices from the three suppliers often still relied on 1y- or 1z-class processes rather than the denser 1a nodes, underscoring that process maturity and product validation timelines frequently lagged pure node announcements.

Summary of Competitive Positioning at the 1y Stage

Supplier	Approximate 1y Volume Start	Primary Early Products	Noted Productivity Claim vs. Prior Node	Strategic Emphasis
Samsung	November 2017	DDR4, LPDDR4X, later LPDDR5	~30 %	Mobile density & volume leadership
SK Hynix	Q1 2019 (after Nov 2018 announcement)	8 Gb DDR4, 16 Gb DDR5	~20 %	DDR4 performance + early DDR5
Micron	2H 2018	12 Gb monolithic LPDDR4x	Not publicly quantified in same terms	Mobile capacity & power

In the competitive landscape of the late 2010s, SK Hynix’s 1y node allowed the company to remain a strong number-two player, close the gap on certain product fronts (notably DDR5 readiness), and improve cost structure relative to its own prior generation. Samsung’s earlier ramp provided a temporary edge in volume and mobile density, while Micron emphasized differentiated mobile offerings. The node illustrated that in DRAM, process timing, product mix, and actual manufacturing efficiency often mattered more than the shared generational label itself. Subsequent nodes would intensify the race around EUV adoption, HBM specialization, and further cell-architecture innovation, but the 1y era remains a clear example of staggered yet closely contested progress among the three leading suppliers.

---

4) Competitive Landscape and Industry Timing

The core topic examines how SK Hynix’s second-generation 10 nm-class (1y / 1Ynm) DRAM process technology positioned the company relative to Samsung Electronics and Micron Technology during the late-2010s transition deeper into the 10 nm class. It also covers the staggered timelines of process ramps, differences in actual manufacturing outcomes, and the product-segment strategies each supplier pursued around 2017–2019. Although all three used the shared “1y” label, real differences in timing, die productivity, and product focus created measurable competitive distinctions.

Relative Timing of Process and Product Launches

Samsung moved first. The company began high-volume production of second-generation 10 nm-class (1y) 8 Gb DDR4 DRAM in November 2017 and quickly extended the process to LPDDR4X and, later, early LPDDR5 packages. By early 2019 Samsung was already shipping high-capacity mobile modules (including 12 GB LPDDR4X) built on 1y-nm dies.

SK Hynix publicly announced development of its 1Ynm 8 Gb DDR4 and the industry’s first JEDEC-compliant 16 Gb DDR5 in November 2018. Initial DDR4 shipments were targeted for the first quarter of 2019, with broader expansion into server, PC, and eventually mobile applications. The same process served as the foundation for SK Hynix’s early commercial DDR5 ramp that began in 2020.

Micron guided for initial 1Y-nm production in the second half of 2018. In November 2018 it announced mass production of the industry’s first monolithic 12 Gb LPDDR4x on the 1Y-nm process, focusing on mobile capacity and power efficiency while continuing to ramp residual 1X output.

Samsung therefore held a roughly one-year lead in volume production. SK Hynix and Micron entered the node in late 2018 with overlapping but differentiated priorities.

Productivity, Die Size, and Cost Differences

Even under the same generational label, the three suppliers delivered different net-die results. Contemporary industry analysis indicated that Samsung’s productivity improvement versus its own 1x generation was the largest (claimed near 30 percent in some reports), giving it the highest number of usable dies per wafer. SK Hynix reported an approximate 20 percent productivity gain for its 1Ynm 8 Gb DDR4 relative to 1Xnm. Micron’s early 1Y position was generally viewed as trailing the two Korean suppliers in net dies, consistent with historical patterns in which node names mask variations in actual design rules, cell size, and yield maturity.

These differences translated directly into cost-per-bit competitiveness once yields stabilized. Higher net dies lowered manufacturing cost and supported stronger margins. Samsung’s earlier ramp therefore conferred a temporary volume and cost advantage, particularly in mobile DRAM, while SK Hynix used its 1y process both to improve its own cost structure and to secure a visible early position in DDR5.

Product-Segment Strategies

• Samsung emphasized mobile density, using 1y-nm dies to enable higher-capacity LPDDR4X and LPDDR5 packages that supported flagship smartphone features (multi-camera systems, larger displays, AI workloads). It maintained parallel strength in server and graphics DRAM.
• SK Hynix applied the node to mainstream DDR4 (highlighting 3,200 Mbps speed and >15 percent power reduction) while simultaneously launching the first JEDEC-compliant DDR5 device. This dual focus allowed it to claim leadership in the emerging DDR5 standard even if overall wafer starts lagged Samsung’s.
• Micron concentrated early 1Y volume on high-capacity monolithic mobile LPDDR4x (12 Gb), stressing power savings of up to 10 percent at 4,266 Mbps data rates while managing a concurrent 1X ramp for broader DDR4 coverage.

Because “first-to-market” claims were often product-specific rather than pure process-node races, SK Hynix’s early DDR5 announcement carried strategic weight beyond pure wafer-start volume.

Industry Context and the Transition Beyond 1y

The 1y generation unfolded against a backdrop of robust DRAM demand driven by server expansion, rising smartphone content, and the early stages of data-center and AI workloads. All three suppliers were simultaneously managing residual 20 nm-class and 1x output while investing in the more complex multi-patterning required for 1y and the forthcoming 1z node.

By 2019–2020 attention had already shifted toward the third-generation 10 nm-class (1z) process. Samsung began 1z mass production in 2019; SK Hynix announced 1Znm 16 Gb DDR4 development later that year; Micron advanced its own 1Z roadmap. Early commercial DDR5 devices from all three suppliers frequently still relied on 1y- or 1z-class processes rather than denser 1a nodes, illustrating that product validation and yield maturity often lagged pure process announcements.

Summary Comparison at the 1y Stage

Supplier	Approximate Volume Start	Primary Early Products	Reported Productivity Gain vs. Prior Node	Strategic Emphasis
Samsung	November 2017	DDR4, LPDDR4X, early LPDDR5	~30 %	Mobile density & overall volume lead
SK Hynix	Q1 2019 (Nov 2018 announcement)	8 Gb DDR4, 16 Gb DDR5	~20 %	DDR4 efficiency + early DDR5 leadership
Micron	2H 2018	12 Gb monolithic LPDDR4x	Not quantified in identical terms	Mobile capacity & power

In the competitive landscape of the late 2010s, SK Hynix’s 1y node enabled the company to remain a strong second-place player, narrow gaps on selected product fronts (notably DDR5 readiness), and improve its internal cost structure. Samsung’s earlier ramp provided a temporary edge in volume and mobile density, while Micron differentiated through high-capacity mobile offerings. The episode underscores a recurring DRAM industry pattern: shared generational labels mask meaningful differences in timing, actual process execution, and product mix—differences that ultimately shape market share and profitability more than the node name itself.

---

5) Broader Implications for DRAM Evolution from SK Hynix’s 1y Node

The core topic explores the longer-term significance of SK Hynix’s second-generation 10 nm-class (1y / 1Ynm) DRAM process technology within the industry’s overall trajectory of scaling, architecture, manufacturing strategy, and product innovation. Far from an isolated process step, the 1y node illustrated key inflection points that continue to shape DRAM development into the mid-2020s and beyond: the practical limits of conventional 6F² planar cell scaling, the rising cost and complexity of multi-patterning, the strategic value of early next-generation product readiness, and the eventual necessity of structural changes such as 4F² vertical-gate architectures and 3D DRAM.

Diminishing Returns of Pure Feature-Size Scaling

By the time the industry reached the mid-10 nm-class nodes, linear shrink factors between successive generations were already rising (often exceeding 0.9), meaning each new node delivered progressively smaller density gains relative to the engineering effort required. Analyses of design-rule trends across Samsung, SK Hynix, and Micron showed that maintaining adequate cell capacitance (typically targeted above 6–7 fF per cell) while shrinking the storage node and transistor became increasingly difficult. Capacitor structures evolved from cylindrical toward quasi-cylindrical designs, high-k dielectrics were thinned and optimized, and transistor reliability features (such as recessed-channel sense-amplifier structures at SK Hynix) were refined.

The 1y generation therefore served as an early signal that pure planar 6F² scaling would not continue indefinitely. Later benchmarking indicated that the conventional 6F² architecture was approaching practical saturation near the low-10 nm regime (around the 1c/1d generations), prompting industry roadmaps to prepare for 4F² cells with vertical-channel transistors and, further out, true 3D DRAM stacking.

Rising Process Complexity and the Shift Toward EUV

At 1y, manufacturers still relied primarily on 193 nm immersion multi-patterning (SADP, LELE, and related techniques). The partial EUV learning that SK Hynix conducted on certain 1ynm layers proved valuable when the company expanded EUV adoption for the 1a node in 2021 and subsequent generations. As patterning steps multiplied at tighter pitches, the economic and yield arguments for EUV strengthened. By the mid-2020s, mature multi-layer EUV strategies had become a competitive differentiator, enabling better pattern fidelity, reduced overlay error, and higher net dies—capabilities that underpinned advanced HBM base dies and high-speed DDR5/LPDDR products.

This transition also highlighted a broader industry pattern: process complexity and capital intensity rise faster than pure density gains once multi-patterning reaches its limits. Suppliers without ready access to EUV tools face steeper challenges in matching leading-edge cost and performance.

Circuit Innovation and Reliability Engineering as Equal Partners to Lithography

The 1y node underscored that continued scaling depends as much on circuit and materials advances as on lithography. SK Hynix’s claimed 4-phase clocking and Sense Amp. Control techniques addressed speed, stability, and power/error issues that intensify with smaller geometries. Similar themes—sensing-margin improvement, row-hammer mitigation, high-k metal-gate adoption in mobile DRAM, and work-function engineering—have remained central in later nodes. These innovations allow manufacturers to extract more usable performance and reliability from a given physical shrink, extending the life of each process generation.

Product Roadmap Leverage and Market Timing

Applying the 1y process simultaneously to high-volume DDR4 and the industry’s first JEDEC-compliant DDR5 demonstrated the commercial value of process reuse. Early DDR5 readiness positioned SK Hynix advantageously when server and data-center demand accelerated. In retrospect, this dual-product strategy foreshadowed later tactics in which a single advanced node supports both commodity DDR/LPDDR and specialized high-bandwidth products (HBM). The ability to validate a new process on mainstream products while preparing next-generation interfaces remains a recurring competitive advantage.

Path Toward Architectural Inflection Points

The challenges visible at 1y—contact margins, external resistance, capacitance retention, and patterning cost—directly informed the industry’s longer-term architectural roadmap. By the mid-2020s, suppliers were actively developing or demonstrating:

• 4F² cell layouts with vertical-gate or vertical-channel transistors to increase density without proportional feature-size reduction.
• Hybrid bonding / CMOS-under-array or wafer-bonding approaches that separate array and periphery processing.
• Materials candidates with higher dielectric constants (for example, STO-based capacitors) and alternative channel materials.
• Eventual 3D DRAM stacking as a longer-term solution once planar scaling economics degrade further.

SK Hynix itself has publicly framed 4F² vertical-gate platforms and 3D DRAM as key pillars for nodes at 10 nm and below, emphasizing continuous innovation in structure, materials, and components to sustain scaling for decades.

Industry Structure and Competitive Dynamics

The staggered 1y ramps among Samsung, SK Hynix, and Micron illustrated that node labels alone do not determine competitive position; timing of volume production, actual net-die productivity, and product mix matter more. Suppliers that convert process gains into both cost reduction and timely next-generation products capture disproportionate share in high-growth segments. In the subsequent AI-driven HBM era, process maturity and packaging capability have become even more decisive, building on the manufacturing discipline and yield focus that nodes such as 1y helped instill.

Enduring Lessons

The 1y DRAM node stands as a representative mid-course checkpoint in DRAM’s long evolution from micron-scale devices to atomic-scale challenges. It showed that:

• Feature-size reduction alone is insufficient; materials, circuit design, and process integration must co-evolve.
• Multi-patterning complexity eventually favors shorter-wavelength lithography and new cell architectures.
• Process investments deliver the highest return when applied across both current-volume and next-generation products.
• Sustainable scaling requires continuous architectural innovation beyond traditional 6F² planar cells.

These themes remain central as the industry navigates the later 10 nm-class nodes, the transition to 4F² designs, and the longer-term prospect of 3D DRAM. The lessons of the 1y generation continue to inform how manufacturers balance density, cost, power, performance, and manufacturability in an era of intensifying demand for high-bandwidth, high-capacity memory.

---