Field report W-97 / L7 · sheet /strategy
Corporate StrategyFiled from the territory
IEDM 2026 Preview: TSMC A14, Ferroelectric NAND and Optical Engines
TSMC targets 2028 volume production for A14 NanoFlex Pro, SK hynix cuts NAND read energy 32 percent, and Samsung tests optical engines on production lines — IEDM 2026 previews them all.
- Grid
- W-97 / L7
- Surveyed
- November 16, 2026
- Reading
- 4 min read
- Correspondent
- Olivia Hart

Field notes
- TSMC's A14 CMOS NanoFlex Pro platform is on track for volume production in 2028.
- IEDM 2026 runs December 12–16 in San Francisco under the theme 'Devices at the Heart of the Intelligence Revolution'.
- SK hynix will show a 176-layer ferroelectric NAND chip; cutting read pass voltage from 6V to 3V could lower read energy per bit by about 32 percent.
- Imec validated high-NA EUV single-exposure patterning for ruthenium interconnects at an 18-nanometer pitch.
- The University of Tokyo's IntMOSAIC demonstrated a 48-die memory stack using organic interposers without through-silicon vias.
TSMC says its A14 CMOS NanoFlex Pro platform is on track for volume production in 2028, and it will headline the 2026 IEEE International Electron Devices Meeting in San Francisco on December 12–16. The conference theme, "Devices at the Heart of the Intelligence Revolution," frames a technical program that now spans transistors, memory, optical interconnect and packaging rather than any single scaling trick.
For practitioners, the tipsheet matters because it shows where the industry is placing its bets. Engineers are opening fresh options in materials, manufacturing, packaging and system design, and several of the demonstrations carry concrete numbers.
What does TSMC's A14 platform change?
A14 CMOS NanoFlex Pro targets mobile processors and demanding AI and computing workloads. The platform couples advances in nanosheet transistors with supporting interconnect and integration technologies, rather than shrinking features alone.
That coordination is the point. The company argues that pairing device innovation with interconnect and integration work is what translates laboratory gains into practical product improvements — the step where previous nodes often stalled.
Why does Intel's germanium work matter?
Intel researchers will present a germanium gate stack built to solve a stubborn trade-off: holding strong electrical performance while keeping insulating layers reliable as they become extremely thin.
The reported combination of mobility, scaling and reliability supports a route toward future architectures that stack complementary transistors vertically. For chip architects, that strengthens the case that computing efficiency can keep improving even after planar scaling loses steam.
Can high-NA EUV simplify manufacturing?
Imec will present electrical validation of high numerical aperture extreme ultraviolet lithography using single exposure patterning for ruthenium interconnects at an 18-nanometer pitch. Single-exposure patterning would cut manufacturing complexity while supporting denser wiring.
The economics are the story. Useful scaling depends on making intricate structures consistently and cheaply, and simplifying the patterning step attacks that constraint directly.
How much energy can new memory save?
Memory research carries some of the hardest numbers in the program:
- SK hynix will demonstrate a 176-layer ferroelectric NAND chip built on a polysilicon channel structure compatible with established NAND manufacturing. The tipsheet projects that cutting read pass voltage from six volts to three would lower read energy per bit by roughly 32 percent.
- The University of Tokyo will present IntMOSAIC, a memory stacking technology using organic interposers. Its demonstrated 48-die stack avoids demanding integration techniques including through-silicon vias, while supporting substantial capacity and bandwidth.
Both approaches matter for AI systems that move enormous quantities of data. The Tokyo work in particular suggests an economical route to placing more memory closer to AI processors — rethinking physical arrangement as an alternative to transistor-level gains.
Is optical interconnect getting production-ready?
Samsung's optical engine integrates electronic and photonic components and uses multiple wavelengths to carry data between chips. The component technologies were fabricated, assembled and tested on production-grade lines, according to the tipsheet.
That manufacturing grounding distinguishes the work from lab-only photonics demonstrations and supports efforts to make optical connectivity a standard building block for AI infrastructure.
What about quantum, cooling and reliability?
A focus session on semiconductor technologies for quantum computing will examine fabrication, control electronics, packaging and yield — the unglamorous steps that determine whether larger quantum systems can actually be built. Applying established semiconductor engineering methods to quantum hardware gives researchers a broader toolkit for integration challenges.
Two reliability contributions round out the program:
- KAIST combines power delivery innovation with cooling and capacitor engineering.
- Imec investigates reliability in atomically thin transistor channels.
Both address the conditions that decide whether promising devices become dependable, shippable technologies.
What should decision-makers take from this?
The research sits at different stages of maturity, and every demonstration in the program needs further development before widespread deployment. But the direction is clear: across companies, universities and research institutes, semiconductor innovation is becoming more connected and more resourceful.
The collective message for planners is that AI's demand curve is spurring practical invention across the entire stack rather than exhausting a single path. IEDM 2026, running December 12–16, will show that momentum in detail — and the 2028 date attached to TSMC's A14 gives the industry its nearest concrete milestone to price into roadmaps.
via semiwiki.com (Source)
More from Olivia Hart
Show full bio
News editor covering regulation, policy and their impact on corporate strategy and chief-executive leadership at CEO Frontier.
456 articles
Adjacent sheets
- PDF Solutions Launches Exensio Aurora, Claims 25x Faster Chip Analytics
- Intel Lands First 18A-P Customer as Microsoft Uses EMIB
- DeepX to Spend Over $100 Million on Samsung-Made AI Chip for Robots
- Memory shortage may tighten through mid-2027, delaying price relief
- Arteris Passes 1,000 Design Starts as AI Reshapes Chip Interconnect