Yipeng Liu, Technical Marketing Director
CDNLive 2015
Taiwan
August 11, 2015
Why Scalability for IoT is Critical to Success
2 © 2015 Cadence Design Systems, Inc. All rights reserved.
• IoT consists of distributed
services based on sensing,
sharing, and controlling
through new nodes
• The nodes are the big
hardware opportunity
– Low cost
– Low energy
– Unique form factors
– Lots of new sensors
– Always alert
• Always alert node may be
independent device or
subsystem in SoC
IoT – The incredible opportunity for innovation
Cloud
Things
Fog
3 © 2015 Cadence Design Systems, Inc. All rights reserved.
Layered cognition
Quiescent system
nW
Active Power
Always Alert
Always On
Application Filter
Cloud Server Farm
Always Alert -detect activity of interest
1-1000 uW
“Hello audio”
Subsystem event processing
1mW - 50 mW
“Turn up volume”
System application processing
50mW-1000mW
“Play ‘Little Apple’”
Cloud application processing
>> 1Watt
“What’s the meaning of life?”
4 © 2015 Cadence Design Systems, Inc. All rights reserved.
The “Things” in IoT
• Characteristics of “Thing”?– A physical entity (node)
– Has one to many sensors
– Connected to the Fog
– Range from very simple to very complicated
– Some can only be powered by batteries
– Most are “always alert”
• Some examples of things– Thermostat, refrigerators, lights, lamppost, smart meter
– Wearable devices: fitness band, watch
• We will focus on the complex, always alert devices
5 © 2015 Cadence Design Systems, Inc. All rights reserved.
• Signal processing workload is growing in complexity over time
• Every IoT application is slightly different
• Energy efficiency is extremely important
• Microcontrollers have limited DSP capabilities
• Cost and ease of use drive desire for one processor core
• No processor cores are ideally suited for all essential functions
– Sensing
– Computation
– Communication
Key design challenges for always alert IoT devices
6 © 2015 Cadence Design Systems, Inc. All rights reserved.
Tensilica Fusion DSP
7 © 2015 Cadence Design Systems, Inc. All rights reserved.
Introducing Tensilica Fusion DSPUltra-low power processing for always on/wearables/IoT
• Sensor fusion
• Biometric monitoring
• Pedestrian dead reckoning
Sensing
• Voice trigger
• Face trigger
• Gesture trigger
Wake-up Processing
• Speech recognition
• Speech pre-processing
• Audio playback
Audio/Voice/ Speech
• Bluetooth Low Energy
• Wi-Fi
• Zigbee
Communications
8 © 2015 Cadence Design Systems, Inc. All rights reserved.
• Base architecture– Built on Xtensa® technology
– Dual-issue, VLIW processor
– Based on HiFi 3 architecture
• Configuration options – Single-precision FPU
• Speeds S/W porting
– AVS (audio/voice/speech)• SW compatibility with HiFi 3 Audio DSP
• Access to 150+ HiFi audio/voice software packages
– 16-bit Quad MAC• Accelerates communications standards like BLE/Wi-Fi
• Accelerates voice algorithm performance
– BLE/Wi-Fi AES-128 • Encryption acceleration for wireless
– Advanced bit manipulation• Accelerates performance of bit operations for
implementation of baseband MAC/PHY
Tensilica Fusion DSP architecture
Xtensa Core ISA
Dual Instruction Issue
Single/Dual MAC DSP
(32b, 24b, 16b)
FPU AVSAES-
128
16-bit
Quad MACAdv. Bit
Manipulation
9 © 2015 Cadence Design Systems, Inc. All rights reserved.
Tensilica Fusion usage scenarios16-bit
Quad
MAC
FPU AVS AES Adv. Bit
Ops
Featu
res
Voice Activation
Sensor Fusion
Audio/Voice/
Speech
Wireless Comms
En
d
Pro
du
cts
Activity Band/
Watch/Phone
Healthcare Band
Smart Home
???
Automotive
10 © 2015 Cadence Design Systems, Inc. All rights reserved.
Tensilica Fusion benchmarks
0
2
4
6
16-bit fixed-point FIR inner
loop
32-bit FP FIR(100 taps x 256
rea;)
MP3 decode 32-bit FP FFT(256 complex)
6-axispedestriannavigation
sensor fusion
Rela
tive P
erf
orm
ance
(Avera
ge R
ISC
= 1
.0)
Superior DSP Performance
RISC A RISC B
Fusion DSP
0
1
2
3
4
0 1 2D
hry
sto
ne
2.1
pe
r M
Hz, w
ith
IP
A
Dhrystone 2.1 per MHz, no IPA, no in-lining
Decent Control Capability
11 © 2015 Cadence Design Systems, Inc. All rights reserved.
Tensilica Fusion SW support from Cadence and partners
DSP Math Libraries (Fixed and Floating Point)
Voice Codecs
Audio Codecs
DSP Audio Framework
Face Trigger
Sensor processing SW
Voice Trigger
Voice Recognition
Voice Pre-Processing
Audio Pre and Post Processing
© 2015 Cadence Design Systems, Inc. All rights reserved worldwide. Cadence and the
Cadence logo are trademarks of Cadence Design Systems, Inc. in the United States and other
countries. All other trademarks are the property of their respective owners and are not affiliated
with Cadence.
12 © 2015 Cadence Design Systems, Inc. All rights reserved.
• Most energy efficient for always on such as Sensory Truly HandsFree Voice Trigger
• Efficient floating-point support for sensor fusion
• Write in C/C++ using advanced compiler with auto-vectorization
• Based upon the Xtensa® 32-bit RISC processor
Leading low-power DSP performance and control
• Multiple configuration options to fit the targeted application
• No waste, maximum efficiency
Configurable - get the right core immediately
• Compatible with HiFi. 70+ partners. 150+ SW packages
• Optimized DSP library with fixed and float kernels
• Cycle-accurate ISS and models for all simulation levels
Comprehensive tools, software, and ecosystem
Proven Xtensa architecture with billions of cores shipped
Fusion DSP benefitsIdeal solution for
IoT, wearable, and
always on functions
© 2015 Cadence Design Systems, Inc. All rights reserved worldwide. Cadence, the Cadence logo, Tensilica, and Xtensa are registered trademarks of Cadence Design Systems, Inc. in
the United States and other countries. All other trademarks are the property of their respective owners.