As complex SoCs continue to evolve towards platforms capable of supporting thousands of applications, so is the complexity of architecting, testing and assuring the multitude of hardware state changes on the SoC that grow at an exponential rate because of the multicore nature of SoCs and the programmable processors used on them.
The paper submitted and this presentation focuses on the growing challenges of management boot and reset sequencing, error recovery, and even power and security management which we combined to call system management, and how system management can be optimized under very complex operations.
2. Today’s Performance SoCs Require
Complex System Management
Samsung Apps Processor (iPhone3) OMAP 4
4+ Compute Engines 4-5 Memory Interfaces
More than 1000 Applications
How Do You Guarantee Proper
Hardware State Changes for +1000 Applications?
3. System Management Complexity
Has Past a Key Threshold
Multiple Application Limited System Management for Multi-
Sequencing sequencing Means Complex System
2010 (Complex (aka random) Operation Verification and
H/W state changes) High Risk Cost/TTM Tradeoffs
When Using Traditional Methods
Single Application
2005 Sequencing
(Simple H/W State Changes)
2000 Interconnect/Dataflow
(Performance Testing)
IP Block (functional test)
1995
SoC System Operation 10X 100X 1000X
Verification Tasks System Operation Verification Complexity
Source: Advance Tech Marketing (Logarithmic)
4. High Risk Cost/TTM Tradeoffs Now Driven
By Inadequate SoC Architecture
• Traditional (Host CPU acts as System Manager)
Host IP
CPU Core
Host CPU no longer has adequate visibility and control
For complex Multicore SoCs
IP
I/O
Core
• Interconnect Based (Adding Control to Data Flows)
Host IP IP
CPU Core Core
Control functions complicate global interconnect design,
experience execution delays, and degrades efficiency
IP
I/O
IP Of high performance data flows
Core Core
These Approaches Were Not Designed to Manage
1000+ Applications Using the Same Device
5. Control Plan Architectures
Reduce Complexity and Ease Tradeoffs
• Separation of control and data planes
make BOTH more efficient CPU DSP
Media
Engine
• Independent controller has System
Controller Global Interconnect
adequate visibility for effective
control Low
Speed
DRAM
High
Speed
Controller
I/O I/O
• Programmable state management
Control Plane
delivers new levels of software/hardware
coupling and implementation flexibility
required for high numbers of applications
• Subsystem IP approach delivers new abstraction layers that
enables system management development acceleration and high
reuse
6. Semico Forecasts High System
Management IP Growth
$180 $1,000
$160 $900
$800
$140
$700
$120
$600
Total M Dollars
M Dollars
$100
$500
$80
$400
$60
$300
$40
$200
$20 $100
$0 $0
2010* 2011* 2012* 2013* 2014* 2015*
Multi Media System Resource Mgmt. Memory Security Computing Communications Video Total
7. Transforming SoC Architecture Using
System Management Subsystem IP
• Complexity levels now dictate that System Management
become an architecture element not an implementation
afterthought
• Control plane architectures introduce new flexibility that
improves system management effectiveness and promote
high reuse
• Subsystem IP approach introduces the abstraction layers
needed to cost effectively manage the system operation
complexity explosion
Make vs. Buy Rapidly Favoring Outsourcing