Solving the SoC Design Dilemma for IoT Applications with Embedded FPGA Tim Saxe, CTO QuickLogic 2017-03-16 Copyright © 2017 QuickLogic, Inc. All rights reserved.
QuickLogic Maximizing Battery Life for Immersive User Experiences in Smartphone, Wearable and IoT devices. Copyright © 2017 QuickLogic, Inc. All rights reserved. 9/21/2018
After 40 years of Higher performance More integration More software Smaller geometries Higher performance More integration More software A new path emerges… Copyright © 2017 QuickLogic, Inc. All rights reserved. 9/21/2018
What Drove the Last 40 Years? Integrated circuits challenged the assumption that a computer was a room sized piece of equipment, And Moore’s Law challenged the assumption that computers were expensive
Old way: Up and to the Right FPGA Software Perspective A CPU for every need Need more performance? Bigger CPU, faster clock and more memory Need even more performance? Multiple CPUs Still not enough performance? Add a hardware accelerator A15 A8 A5 R7 Performance R4 M4 M3 M0 Capability
What is Changing Now? Mask costs Fragmented markets The role of software Owning the BOM
Mask Costs In the 80’s, masks were cheap and transistors were expensive Adding enough eFPGA to be useful was prohibitively costly At smaller geometries the cost of adding a useful amount of eFPGA gates becomes reasonable and the cost of a spin becomes really painful
Fragmented, Changing Markets Most IoT markets are very new: The good news is that means opportunity The bad news is that makes it very difficult to tell what they need Need multiple SKUs “Follow a build-test-learn process.” HBR “Components must be … amenable to … still-unidentified future products.”, McKinsey
Build-Test-Learn We’ve been helping people do this for nearly 30 years (used to be called burn and learn) Easy to do with eFPGA, really hard and expensive with silicon Goes against the “Do It Right The First Time” semiconductor culture
What about ‘still-unidentified future products’? “How do I design for something that I don’t know?” ‘Want Some Creativity? Crank-up the Constraints’ Bob Sutton d.school Be amazed what can be done with a small amount of programmable logic
When Software is not the answer CPU too slow? Use an FPGA Dedicated hardware can easily give a 5x to 10x performance advantage CPU too power hungry? Use an FPGA Dedicated hardware can easily give a 5x to 10x power advantage Battery Life M0 M3 M4 R4 R7 A5 A8 A15 Capability FPGA Performance M0 M3 M4 R4 R7 A5 A8 Capability FPGA
‘Components must be small enough to be embedded in today’s smart watches and smart glasses but also amenable to further shrinking’ -- McKinsey Every extra component takes up volume that could be used to add battery capacity, or make the product smaller – eFPGA eliminates a component
Every extra component brings another vendor into the equation The more complete the solution, the fewer potential competitors and a larger percentage of the available BOM
eFPGA Summary Transistor costs now low enough that a useful amount of eFPGA costs a few cents Mask costs now high enough that adding a few cents to die cost to save a design spin has a good ROI Rapidly evolving market favors fast TTM eFPGA solutions Fragmented market combined with high mask costs favor flexible solutions When performance is critical, hardware beats software When power is critical, hardware beats software Tools, not silicon, are the key to FPGA success
Un-named, 2nd Top-Tier Foundry Secured Two Top-tier eFPGA Foundries 1st Embedded FPGA IP for GLOBALFOUNDRIES’ 22FDX® ArcticPro™, first eFPGA announced for new GLOBALFOUNDRIES’ 22FDX® (FD-SOI) process Second top-tier foundry to port, and market for licensing, that expands the availability to semiconductor companies and OEMs developing new designs targeting this foundry's high volume 40nm fabrication process Design proven in 65nm and 40nm ArcticLink and EOS products that are used in a variety of commercial products Un-named, 2nd Top-Tier Foundry Copyright © 2017 QuickLogic, Inc. All rights reserved. 9/21/2018 15