📈 Yield Recovery and Countermeasures in PSRAM


🧠 Background

PSRAM was developed in 2001 by reusing a 0.25 µm DRAM base process to address the emerging mobile phone market, which demanded:

To secure early market entry—most notably for the world’s first camera-equipped mobile phones—mass production was launched despite an extremely low initial yield of ~30%.

Yield recovery therefore had to be executed in parallel with volume production, under intense schedule and business pressure.

At the same time, the company’s long-term strategy was already shifting toward LCD driver ICs and high-voltage mixed CMOS, placing strict limits on:


🚧 Constraints

Category Constraint
Process No major redesign allowed (DRAM base fixed)
Schedule Extremely short recovery timeline
Business Market entry prioritized over yield maturity
Reliability 90 °C guaranteed operation required
Strategy Limited long-term commitment to memory

These constraints defined the solution space for yield recovery.


⚠️ Dominant Failure Modes

Two DRAM-derived failure modes became critical under mobile operating conditions.

⏸ Pause Refresh Fail


🔁 Disturb Refresh Fail

📌 Both modes were minor or latent in conventional DRAM, but were amplified by mobile usage and temperature requirements.


⚛️ Physical Root Causes

⏸ Pause Failure Mechanism

Cell retention time is governed by:

\[\tau = \frac{C_{cell} \cdot V_{cell}}{I_{leak}}\]
Parameter Status
Ccell Met design target
Vcell Met operating spec
Capacitor integrity Confirmed
Dominant limiter Junction leakage (Ileak)

Leakage current increased exponentially with temperature, indicating process-induced junction damage.

Root cause analysis linked this behavior to accumulated plasma exposure in the DRAM process flow.


🔁 Disturb Failure Mechanism

Observed characteristics:

📌 This identified critical CD control of cell transistors as essential for suppressing disturb failures.


🛠 Yield Recovery Countermeasures

Yield recovery required multi-layer optimization without stopping mass production.


⚙️ Process-Side Measures

Measure Intent
Plasma exposure elimination Reduce junction defect density
O₂ plasma → wet stripping Suppress plasma-induced damage
HF clean count minimization Preserve gate oxide integrity
Contact damage suppression Reduce leakage paths

🎯 Focus: remove physical damage, not mask symptoms.


🧱 Device-Side Measures

Measure Effect
Back-bias (−1 V → −3 V) Reduced junction leakage, higher effective Vth
Critical CD tightening Suppressed SCE-induced disturb
Channel length margin enforcement Stabilized Ioff

🔄 Operation-Side Measures

📌 All measures were executed through cross-functional coordination between process, device, and manufacturing teams.


✅ Result

Metric Outcome
Initial yield ~30%
Recovered yield Stable 80% class
Reliability Pause / Disturb largely suppressed
Production Continuous, no line stop
Market impact Enabled early mobile camera phones

From a purely technical standpoint, the yield recovery was successful and met immediate market needs.


🧭 Strategic Outcome

Despite technical recovery:

Evaluation of a 0.18 µm trench DRAM confirmed that high-temperature leakage was structural, not incidental.

At the same time, corporate strategy had converged on:

Sustaining mobile memory would have required:

— commitments incompatible with this strategic direction.

As a result, DRAM-derived mobile memory development was strategically terminated, and resources were reallocated to logic-oriented and mixed-signal products where existing process assets could be leveraged more effectively.


🧠 Legacy Insight

PSRAM represents a rare but instructive case:

Yield can be recovered, and reliability targets can be met,
yet architectural scalability and business sustainability
may still fail.

This case demonstrates that:

📘 PSRAM stands as a textbook example of
technology pushed to its rational boundary.