High-Performance Memories for Packet Processing

High-Performance Memories
for Packet Processing
Chris Johnson
Global DRAM Applications Engineering Manager
©2012 Micron Technology, Inc. All rights reserved. Products are warranted only to meet Micron’s production data sheet specifications. Information, products, and/or specifications
are subject to change without notice. All information is provided on an “AS IS” basis without warranties of any kind. Dates are estimates only. Drawings are not to scale. Micron and
the Micron logo are trademarks of Micron Technology, Inc. All other trademarks are the property of their respective owners.
©2012 Micron Technology, Inc.
|
1
High Performance Networking Memories
Requirements
•
•
•
•
•
•
•
Low-tRC/high access rate
High bandwidth
Density and Scalability
Low pin-count
Small footprint
Low power consumption
Reasonable system cost
Trends
•
•
•
•
•
Low-tRC architectures
Managed DRAM
SerDes interfacing
2.5/3D integration
Photonics
Requires Leading Edge DRAM Technologies
Commodity DRAM Architectures do not address these needs
©2012 Micron Technology, Inc.
|
2
Commodity
DRAMDRAM
Drivers
(DDR1/2/3/4)
Commodity
Drivers
Volume
Die Size

Primary driver

Increased die per wafer

Equates directly to yield

Improved yield

Manufacturing efficiencies
Test times
Tooling

Back end flexibility

Amortized R&D cost
Appropriate Design Margin
IO Count
Commodity Functionality
©2012 Micron Technology, Inc.
|
3
Solutions for High Performance
Reduced Page Size

Faster core timing

Lower power

Increases die size
Beef Up Power Bussing

Eliminates tRRD and tFAW type
restrictions

Adds metal count or die size
Reduce Turn Times

Change functionality or architecture

Added posting registers

Adds die size
©2012 Micron Technology, Inc.
|
4
RL3 Solves Access Rate Gap
Reduce Random Access
(Row Cycle Time

Fragments the array

Reduced digit line length
Bus Turn-around Time

Add WRITE registers

Eliminate internal bus contention
Eliminate RAS Chain Limitations

Move to flip chip

Increased power bussing
©2012 Micron Technology, Inc.
|
5
Bus Turnaround Delay Effect on Bandwidth
RL3 Delivers Sustained Bus Utilization Under Networking Workloads
RL3-2133 BL8
RL2-1066 CIO BL8
DDR3-2133 BL8 CL=13
2,133
Bandwidth per Data Pin (Mb/s)
1,867
1,600
1,333
1,067
800
533
267
0
0.01
0.10
1.00
10.00
100.00
Read/Write Ratio
Key Parameters
• <8ns tRC
• x36/x18 configurations
• tRRD/tFAW – no delay
• Burst Lengths 2,4,8
©2012 Micron Technology, Inc.
|
6
Process Characteristics
DRAM Process

Create really good capacitors
Memory Cells
Limits Speed
Increases Power

Limited Metal Layers
Low Cost
Logic Process

Low Capacitance
High speed
Low Power

Large number of metal steps
Design flexibility
©2012 Micron Technology, Inc.
|
7
Hybrid
Memory
Cube (HMC)
Hybrid
Memory
Cube
Introduction of Managed DRAM
Through-Silicon Vias
(TSV)
Abstraction
Protocol
DRAM
DRAM
DRAM
DRAM
Many
Buses
> 1Tb/s
Processor
Logic Die
High Speed Links
Reduced Pin count
Increased Bandwidth
Note: Tb/s = Terabits / second
©2012 Micron Technology, Inc.
|
8
HMC Architecture
Logic Base
Vault Control
Vault Control
Vault Control
Vault Control
Add advanced switching,
optimized memory control
and simple interface to host
processor(s)…
Memory Control
Crossbar Switch
Link Interface
Controller
Link Interface
Controller
Link Interface
Controller
Processor
Links
HMC
T
X
Link Interface
Controller
3DI & TSV Technology
DRAM7
DRAM6
DRAM5
DRAM4
DRAM3
DRAM2
DRAM1
DRAM0
Logic Chip
Host
16 Lanes
R
X
Vault
R
X
16 Lanes
T
X
DRAM
Logic Base
©2012 Micron Technology, Inc.
|
9
2.5D/2D/MCM System/3D
2.5D
SOC
IP Block
DRAM Stack
Si Interposer(2.5D)
MCM substrate (2D)
▶
DRAM IP Block part of SOC

Can be tailored to SOC interface requirement

Emulate another DRAM technology interface if necessary

Manage DRAM arrays
▶
Reduced memory footprint
▶
Various DRAM stacks for different purposes
(e.g. low-power or low-latency)
▶
Short channel routing
▶
Ease of implementation into foundry model

DRAM IP block available as part of node library
3D Opens New Possibilities
DRAM Stack
e.g. 64GB/s
Parallel Interface
PHY
BIST,
Repair, APG
Memory Control (Refresh
Scheduling, Sequencer, Read/Write
Buffers, etc. )
ASIC
*Size relationship of memory stack, IP blocks, and ASIC for
illustration only and are not to scale.
©2012 Micron Technology, Inc.
|
10