dram scaling 1974 choose between etching down trench building up stacked

# Step 2 — Choose Between Etching Down (Trench) and Building Up (Stacked): Two Geometries for the Third Dimension

## 1. The Fundamental Fork: Substrate Etching vs. Over-Gate Construction

Once the cell capacitor must leave the two-dimensional plane to preserve signal charge, the memory architect faces a definitive structural fork: etch downward into the silicon bulk, or build upward over the access transistor. In Step 1, we showed why planar scaling collapses: as the feature size $F$ contracts, a flat capacitor's surface area falls quadratically ($A \propto F^2$), driving capacitance $C_s$ below the $30\text{--}40\ \text{fF}$ floor required for bit-line sensing margins and soft-error immunity. Because gate oxide thinning ($t_{ox}$) is constrained by Fowler-Nordheim tunneling and dielectric breakdown ($E > 10\ \text{MV/cm}$), the only path to recovering surface area without enlarging the cell footprint is to exploit the vertical dimension ($Z$-axis):

$$A_{\text{planar}} \approx F^2 \quad \longrightarrow \quad A_{\text{3D}} \approx P \cdot h \gg F^2$$

where $P$ is the perimeter of the storage node and $h$ is the vertical height or depth. This transformation decouples capacitor surface area from planar layout area, but it creates two opposing engineering architectures:
1. The Trench Capacitor (Etching Down): A narrow, high-aspect-ratio hole ($3\text{ to }8\ \mu\text{m}$ deep) is etched by deep reactive-ion etching (RIE) directly into the single-crystal silicon substrate. The capacitor dielectric and counter-electrode (poly-Si plate) line the interior of the trench. The wafer surface remains flat, preserving lithographic depth-of-focus, but adjacent trenches risk punch-through and leakage through the silicon bulk.
2. The Stacked Capacitor (Building Up): The access transistor is fabricated first, and the storage capacitor is constructed above the gate electrode in polysilicon and inter-layer dielectrics. Because the storage node is elevated above the substrate, substrate leakage and alpha-particle carrier collection are virtually eliminated, but the resulting vertical topography creates severe step heights that challenge optical lithography and metallization.

Trench vs. Stacked Capacitor: The Two 3D Architectures etching downward into the bulk substrate versus building upward over the access transistor OPTION 1: TRENCH CAPACITOR (ETCH DOWN) Si Substrate (Flat Surface) Access Gate Trench Storage Node Perfect Surface Planarity (h = 0 above plane) Depth h = 4–8 µm into bulk substrate OPTION 2: STACKED CAPACITOR (BUILD UP) Si Substrate Base Gate Stacked Node (Crown/Pillar) Height Δh = 1–2 µm (Severe Step) Isolated above substrate: zero bulk leakage THE ARCHITECTURAL TRADEOFF ✓ Trench: Preserves wafer planarity for lithography, but risks substrate leakage and punch-through deep in the bulk ✓ Stacked: Eliminates substrate leakage and soft errors, but creates extreme topography that challenges optical focus The DRAM industry split between these two camps for over a decade before advanced planarization matured.

## 2. Compare the Failure Modes: Topography vs. Substrate Leakage

Choosing between trench and stacked capacitor technologies is an explicit choice between two entirely different physical and manufacturing failure mechanisms.

The Physical Failure Modes: Bulk Leakage vs. Lithographic Defocus the manufacturing vulnerabilities that defined the trench-versus-stacked competition TRENCH RISK: BULK PUNCH-THROUGH Trench 1 Trench 2 Substrate Punch-Through I_leak across bulk Si Substrate Coupling & Alpha Errors • Trench-to-trench crosstalk as spacing collapses • Alpha particles generate charges in bulk collected by trench • RIE profile control: micro-trenching and bowing STACKED RISK: LITHOGRAPHY STEP HEIGHT Poly Tower Lens Depth-of-Focus (DOF) Loss Topography & Interconnect Opens • 1–2 µm vertical step exceeds optical depth of focus • Metal lines thinning over steps cause open circuits • Required Chemical Mechanical Polishing (CMP) to solve Two Distinct Engineering Philosophies: Trench bet on silicon etching and bulk isolation; Stacked bet on thin-film deposition and CMP planarization.

## 3. Place This Decision in the Project's Own Lineage

Step 2 marks the moment DRAM cell engineering separated into two parallel evolutionary branches:
1. The Extension of the 1968 Dennard 1T1C Topology: The circuit topology invented in 1968 remains strictly preserved: one access transistor connected in series with one storage capacitor. Neither trench nor stacked cells alter the fundamental charge-transfer equation ($\Delta V = \frac{C_s}{C_s + C_{bl}} V_{dd}$). What changes is the physical geometry of the capacitor node, moving from a 2D planar square to a 3D volume.
2. The Industrial Divide: For more than a decade, the semiconductor industry divided along this architectural boundary:
- Trench Advocates (IBM, Texas Instruments, Siemens): Valued surface planarity above all else, preserving lithographic focus margins in the early sub-micron era before planarization tools existed.
- Stacked Advocates (Hitachi, Toshiba, Samsung, Micron): Prioritized electrical isolation, avoiding substrate punch-through and alpha-particle soft errors while pioneering the chemical mechanical polishing (CMP) processes that would later dominate all semiconductor metallization.
3. The Unification in the Third Dimension: Regardless of whether the capacitor dug into the earth or reached into the sky, both paths proved the same central physical principle: to continue scaling memory density, semiconductor manufacturing had to conquer the $Z$-axis.

Step 2 establishes the structural choices of 3D memory; Step 3 will explore how high-aspect-ratio reactive ion etching and conformal dielectric deposition made these deep vertical structures physically manufacturable.

Go deeper with CFSGPT

Get AI-powered deep-dives, save terms, and run advanced simulations — free account.

Create Free Account