LPDDR5 Architecture and Physical Structure¶
LPDDR5 (Low Power Double Data Rate 5) is designed for mobile and power-constrained applications. This guide explores its physical structure and key architectural differences from standard DDR memory.
LPDDR5 Overview¶
LPDDR5 introduces significant architectural changes optimized for:
- Higher bandwidth with per-channel data rates up to 6400 MT/s
- Lower power with voltage scaling and power-down modes
- Improved efficiency with dedicated channels and bank groups
Key Specifications¶
| Parameter | LPDDR5 | LPDDR4X (Comparison) |
|---|---|---|
| Max Data Rate | 6400 MT/s | 4266 MT/s |
| Voltage (VDD2) | 1.05V | 1.1V |
| Voltage (VDDQ) | 0.5V | 0.6V |
| Prefetch | 16n | 16n |
| Burst Length | BL16, BL32 | BL16 |
| Channels | 2 per die | 2 per die |
| I/O Width | x16 per channel | x16 per channel |
Physical Structure Overview¶
Die Organization¶
LPDDR5 organizes memory in a hierarchical structure with independent channels:
┌─────────────────────────────────────────────────────────────┐
│ LPDDR5 Die │
├─────────────────────────────┬───────────────────────────────┤
│ Channel A │ Channel B │
│ │ │
│ ┌──────────────────────┐ │ ┌──────────────────────┐ │
│ │ Bank Group 0 │ │ │ Bank Group 0 │ │
│ │ ┌──────┬──────┐ │ │ │ ┌──────┬──────┐ │ │
│ │ │Bank 0│Bank 1│ │ │ │ │Bank 0│Bank 1│ │ │
│ │ └──────┴──────┘ │ │ │ └──────┴──────┘ │ │
│ │ ┌──────┬──────┐ │ │ │ ┌──────┬──────┐ │ │
│ │ │Bank 2│Bank 3│ │ │ │ │Bank 2│Bank 3│ │ │
│ │ └──────┴──────┘ │ │ │ └──────┴──────┘ │ │
│ └──────────────────────┘ │ └──────────────────────┘ │
│ │ │
│ ┌──────────────────────┐ │ ┌──────────────────────┐ │
│ │ Bank Group 1 │ │ │ Bank Group 1 │ │
│ │ ┌──────┬──────┐ │ │ │ ┌──────┬──────┐ │ │
│ │ │Bank 4│Bank 5│ │ │ │ │Bank 4│Bank 5│ │ │
│ │ └──────┴──────┘ │ │ │ └──────┴──────┘ │ │
│ │ ┌──────┬──────┐ │ │ │ ┌──────┬──────┐ │ │
│ │ │Bank 6│Bank 7│ │ │ │ │Bank 6│Bank 7│ │ │
│ │ └──────┴──────┘ │ │ │ └──────┴──────┘ │ │
│ └──────────────────────┘ │ └──────────────────────┘ │
│ │ │
│ ┌──────────────────────┐ │ ┌──────────────────────┐ │
│ │ Bank Group 2 │ │ │ Bank Group 2 │ │
│ │ ... │ │ │ ... │ │
│ └──────────────────────┘ │ └──────────────────────┘ │
│ │ │
│ ┌──────────────────────┐ │ ┌──────────────────────┐ │
│ │ Bank Group 3 │ │ │ Bank Group 3 │ │
│ │ ... │ │ │ ... │ │
│ └──────────────────────┘ │ └──────────────────────┘ │
├─────────────────────────────┼───────────────────────────────┤
│ CA Bus + DQ[15:0] │ CA Bus + DQ[15:0] │
│ x16 I/O │ x16 I/O │
└─────────────────────────────┴───────────────────────────────┘
Channels¶
Dual Channel Architecture¶
LPDDR5 features two independent channels per die:
| Feature | Channel A | Channel B |
|---|---|---|
| Data Width | x16 | x16 |
| CA Bus | Independent | Independent |
| Clock | CK_A | CK_B |
| Operation | Simultaneous with Channel B | Simultaneous with Channel A |
Channel Benefits¶
graph TD
subgraph "Single Channel Access"
SC[Memory Controller] --> CH1[Channel]
CH1 --> BW1[16-bit × Rate]
end
subgraph "Dual Channel Access"
DC[Memory Controller] --> CHA[Channel A]
DC --> CHB[Channel B]
CHA --> BWA[16-bit × Rate]
CHB --> BWB[16-bit × Rate]
end
Total Bandwidth = 2 channels × 16 bits × Data Rate
Example at 6400 MT/s: - Per channel: 16 bits × 6400 MT/s = 12.8 GB/s - Total: 2 × 12.8 GB/s = 25.6 GB/s
Bank Groups and Banks¶
Bank Group Organization¶
LPDDR5 includes 4 bank groups per channel, each containing 4 banks:
| Structure | Count |
|---|---|
| Bank Groups per Channel | 4 |
| Banks per Bank Group | 4 |
| Total Banks per Channel | 16 |
| Total Banks per Die | 32 (2 channels × 16) |
Bank Group Timing Advantages¶
Similar to DDR4, operations to different bank groups have reduced timing constraints:
| Parameter | Same Bank Group | Different Bank Group |
|---|---|---|
| tCCD | tCCD_L (Long) | tCCD_S (Short) |
| Access | Sequential dependency | More parallelism |
I/O Interface¶
Data Bus¶
| Signal | Width | Description |
|---|---|---|
| DQ | x16 per channel | Bidirectional data |
| DQS/DQS# | 2 pairs per channel | Data strobe (differential) |
| DMI | 2 per channel | Data mask/invert |
Command/Address Bus¶
LPDDR5 uses a multiplexed CA bus:
| Signal | Width | Description |
|---|---|---|
| CA | 7 bits | Command/Address |
| CK/CK# | 1 pair | Differential clock |
| CS | 1-2 | Chip select |
| CKE | 1 | Clock enable |
Signaling¶
| Aspect | LPDDR5 |
|---|---|
| CA Signaling | Single-ended CMOS |
| DQ Signaling | POD (Pseudo Open Drain) |
| DQS Signaling | Differential |
| Termination | On-die termination |
Data Strobe Configuration¶
Byte Mode vs x16 Mode¶
LPDDR5 supports flexible DQS mapping:
Byte Mode (x8):
x16 Mode:
Write Clocking Architecture¶
WCK (Write Clock)¶
LPDDR5 introduces a dedicated write clock (WCK):
┌────────────────────────────────────────┐
│ Memory Controller │
│ │
│ CK ──────────────────────────────────┼───▶ DRAM CK
│ │
│ WCK ─────────────────────────────────┼───▶ DRAM WCK
│ │
│ DQ/DQS ◀────────────────────────────┼───▶ DRAM DQ/DQS
│ │
└────────────────────────────────────────┘
CK ─┐ ┌─┐ ┌─┐ ┌─┐ ┌─ Command/Address timing
└─┘ └─┘ └─┘ └─┘
WCK ┐┌┐┌┐┌┐┌┐┌┐┌┐┌┐┌ Data timing (2× or 4× CK)
└┘└┘└┘└┘└┘└┘└┘└┘
WCK Ratios¶
| WCK:CK Ratio | Usage |
|---|---|
| 2:1 | Standard operation |
| 4:1 | High-speed operation |
Ranks¶
Multi-Die Packages¶
LPDDR5 packages can contain multiple dies (ranks):
┌─────────────────────────────────────────────────────────────┐
│ LPDDR5 Package │
├─────────────────────────────────────────────────────────────┤
│ ┌─────────────────────┐ ┌─────────────────────┐ │
│ │ Die 0 │ │ Die 1 │ │
│ │ (Rank 0) │ │ (Rank 1) │ │
│ │ │ │ │ │
│ │ Ch A Ch B │ │ Ch A Ch B │ │
│ └─────────────────────┘ └─────────────────────┘ │
│ │ │ │
│ └───────────┬──────────────┘ │
│ │ │
│ Shared CA/DQ Bus │
└─────────────────────────────────────────────────────────────┘
Rank Selection¶
| CS Pattern | Selected |
|---|---|
| CS0 = L | Rank 0 |
| CS1 = L | Rank 1 |
| Both H | Deselected |
Comparison: LPDDR5 vs DDR4¶
| Feature | LPDDR5 | DDR4 |
|---|---|---|
| Target | Mobile | Server/Desktop |
| Channels/Die | 2 | N/A (module level) |
| I/O per Channel | x16 | x4/x8/x16 |
| Bank Groups | 4 per channel | 4 per device |
| Banks | 16 per channel | 16 per device |
| Prefetch | 16n | 8n |
| Write Clock | Dedicated WCK | Same as CK |
| CA Interface | Multiplexed (7-bit) | Parallel |
| VDD | 1.05V | 1.2V |
| VDDQ | 0.5V | 1.2V |
Power Management¶
Low Power Features¶
LPDDR5 includes extensive power management:
| State | Description | Power |
|---|---|---|
| Active | Normal operation | Full |
| Idle | No commands | Reduced |
| Power Down | Clock gated | Low |
| Self Refresh | Refresh maintained internally | Very Low |
| Deep Sleep | Minimum retention | Lowest |
Deep Sleep Mode (DSM)¶
LPDDR5 introduces Deep Sleep Mode for extreme power savings:
- Data retention maintained
- No clock required
- Longest wake-up time
- Ideal for always-on applications
Addressing¶
Address Mapping¶
| Component | Bits | Description |
|---|---|---|
| Channel | 1 | Select Channel A or B |
| Bank Group | 2 | Select 1 of 4 bank groups |
| Bank | 2 | Select 1 of 4 banks |
| Row | 14-17 | Row address |
| Column | 10 | Column address |
Validation Considerations¶
Key Test Areas¶
- Dual channel operation
- Bank group interleaving
- WCK:CK ratio operation
- Power state transitions
- Deep Sleep entry/exit
- CA bus timing
- POD signaling levels
Common Issues¶
| Issue | Symptom | Debug Approach |
|---|---|---|
| Channel skew | Bandwidth asymmetry | Check per-channel timing |
| WCK alignment | Data errors | Verify WCK-DQS relationship |
| Power state | Corruption on wake | Check refresh timing |
| CA setup/hold | Command errors | Measure CA eye |
Related Topics¶
- LPDDR High-Speed Validation - Signal integrity and timing
- LPDDR Eye Diagram Analysis - Eye measurements
- DDR Architecture - DDR4 comparison
References¶
- JEDEC JESD209-5 LPDDR5 Standard
- Mobile memory technical documentation (publicly available)