`Case 5:18-cv-07581-LHK Document 31-9 Filed 03/04/19 Page 1of5
`
`EXHIBIT 9
`EXHIBIT 9
`
`EXHIBIT 9
`EXHIBIT 9
`
`
`
`®
`
`EFLX embedded FPGA
`
`Case 5:18-cv-07581-LHK Document 31-9 Filed 03/04/19 Page 2 of 5
`
`EFLX® Embedded FPGA
`Dense, Fast, Proven, Scalable
`Everything You Need for an eFPGA in Your SoC
`✔ High density similar to full custom FPGA
`✔ High performance similar to full custom FPGA
`✔ eFPGA IP Core silicon proven
`✔ eFPGA Arrays of any size by tiling proven eFPGA IP cores to the size needed
`✔ Options for DSP and RAM as you need
`✔ Available and silicon proven in TSMC 16, 28, 40 & on demand for any TSMC node
`✔ Compatibility with your metal stack and your voltage range
`✔ Software tools with a Graphical User Interface
`✔ Timing (cid:31)les extracted from design database and available across all corners
`✔ DFT fault coverage of 99% with special logic for 100x acceleration of test time
`
`TSMC IP Alliance Member
`
`Flex Logix® is now a TSMC IP Alliance Member based on the work it has done with TSMC over the
`past several years to develop embedded FPGA IP meeting TSMC’s standards for documentation,
`design methodology, and engineering validation in silicon. Flex Logix will continue to prove all
`EFLX® embedded FPGA IP in silicon after rigorous engineering checks and sign-o(cid:30)s.
`
`TSMC 40/28/16, GF 14, Sandia 180, others in 6 mo.s
`TSMC 16FF+/FFC
`EFLX150 Gen 2
`
`PROVEN IN SILICON
`TSMC 16FF+/FFC
`EFLX4K Gen 2
`
`PROVEN IN SILICON
`TSMC 28HPM/HPC
`EFLX2.5K Gen 1
`
`PROVEN IN SILICON
`TSMC 28HPC/HPC+
`EFLX4K Gen 2
`
`In fabrication
`TSMC 40ULP/LP
`
`EFLX 100 Gen 1
`
`PROVEN IN SILICON
`GF 14LPP
`
`EFLX4K Gen 2
`
`In design
`Sandia 180
`
`EFLX4K Gen 2
`
`In fabrication
`
`Detailed product briefs are available for each EFLX core. Operating temperature range is -40C to +125C Tj.
`Multiple voltage ranges are supported. The EFLX Compiler has timing at numerous corners for each core.
`We an port to other foundries/process nodes in ~6 months.
`Multiple Customers and Markets Adopting eFPGA
`Multiple customers have built chips using EFLX eFPGA and more are in design. Customers include DARPA,
`Harvard, HiPer, Sandia, SiFive & more. Markets and value propositions include:
`
`SoCs, ML, Vision Reconfigurable accelerators & algorithm iteration in real time.
`Networking
`Reconfigurable switches, NICs, etc which can be updated for changing protocols and needs.
`Data Center
`Accelerators and Processor+FPGA integration.
`Base Stations
`Reconfigurable DFE integrated for lower power and higher performance.
`MCU+IoT
`Reconfigurable accelerators & serial I/O; battery life extension
`Aerospace
`Smaller, lighter, lower power & higher performance through integration.
`
`
`
`Case 5:18-cv-07581-LHK Document 31-9 Filed 03/04/19 Page 3 of 5
`
`EFLX Embedded FPGA
`XFLX™, ArrayLinx™ & RAMLinx™ Interconnects
`Revolutionary Interconnect Enables Density and Scalability
`Traditional FPGA fabrics are only 20% programmable logic: the programmable interconnect takes 80% of the area!
`
`Flex Logix has developed revolutionary new interconnects:
`1. XFLX Interconnect: reduces the programmable interconnect area required by about 1/2 AND reduces the number of metal layers required. This enables
`density similar to full custom FPGA and compatibility with most metal stacks.
`2. ArrayLinx Interconnect: this allows a large number of eFPGA array sizes to be built quickly using a single, silicon proven eFPGA IP core.
`3. RAMLinx Interconnect: this allows any type and amount of RAM to be quickly interleaved in an eFPGA array using silicon proven eFPGA IP cores.
`
`These interconnects are covered by numerous issued and pending patents in the USA and other countries.
`Density & Performance Similar to Full Custom FPGA
`The XFLX interconnect takes ~1/2 the area of traditional FPGA mesh interconnect, so even though we use standard cells, for rapid implementation, we
`achieve density and performance similar to full custom FPGA. And we use only 5-7 metals layers, so we are compatible with most metal stacks.
`
`Embedded FPGAs from 150 LUTs to ~200K LUT4s
`EFLX arrays are constructed from one of two building blocks: the EFLX150 core with ~150 LUT4s and the
`EFLX4K core with ~4K LUT4s. Both are complete embedded FPGAs with programmable logic, program-
`mable interconnect, I/O, clock circuitry and configuration logic. They also have a top level ArrayLinx
`Interconnect which is used to build arrays of larger size by “tiling” into arrays with no GDS change. Arrays
`of different sizes can be delivered in days & customers are encouraged to use multiple, different size arrays
`in a single design. The LUTs in our Gen 2 architecture are 6-input LUTs for higher speed and density. All
`EFLX cores are proven in silicon in arrays of at least 2x2 so all array sizes are proven out. If you need even
`larger arrays, ask about our roadmap to ~800K LUT4 arrays.
`Thousands of Interface Pins
`A single EFLX4K core has >1000 interface pins: 632 in and 632 out; larger arrays have much more.
`You can connect EFLX embedded FPGAs into wide, fast buses and wide data and control paths; the
`interfaces are standard CMOS so they run very fast.
`Optional MACs for DSP/AI/ML Acceleration
`Both the EFLX-100/150 and EFLX4K cores are offered in versions where some of the LUTs are replaced
`with Multiplier-Accumulator (MAC) blocks consisting of a 22-bit pre-adder, 22-bit multiplier and 48-bit
`accumulator which can be pipelined for very fast DSP implementations. For AI/ML, the multiplier can be
`configured as two 11x11 multipliers for double the throughput.
`Optional RAM: Any Kind, Any Amount
`The EFLX cores contain memory. If you require more, it can be synthesized at the edge of the array or between the
`cores, within the array. The EFLX Compiler software will map your RTL onto the RAM as part of the array. We synthe-
`size RAM from standard compilers with optional MBIST: single port or dual port, whatever size and number of blocks you
`want, with or without parity/ECC. We can also integrate your special memory like TCAM or custom accelerators.
`Test and Reliability Features
`Test vectors are available with stuck-at fault coverage of 99%. Our new Gen 2 Architecture has a special test mode that accelerates test speeds 100x over
`our first generation. For High Reliability applications we have the ability to read back the configuration bits; and “scrubbing” is also possible to re-write
`configuration bits periodically.
`
`
`
`Case 5:18-cv-07581-LHK Document 31-9 Filed 03/04/19 Page 4 of 5
`
`EFLX Embedded FPGA
`Integration & Programming
`
`Flex Logix has
`interface RTL
`for AXI, AHB
`and APB
`buses as a
`Slave or
`Master.
`
`Design Deliverables
`We deliver the exact EFLX array size you specify, your mix of all-LUT and LUT-DSP cores, and RAM if needed. If you want multiple arrays of different
`sizes/configurations we can easily accommodate that. If your requirements change mid-design we can quickly deliver a larger or smaller array.
`The design files we deliver are: LEF, LIB, Verilog™ Model, SDF, GDS-II™, CDL/Spice netlist, and DFT test vectors (>>98.5% coverage).
`We designed and built a validation chip which is proven in silicon using these same files.
`Power Management Modes & IR Drop Analysis
`EFLX cores can be power-gated to reduce static power if not in use; in TSMC 40ULP we added more power management modes and 0.5V state retention.
`We use EDA tools to analyze the EFLX array for IR drop in static & high-switching activity dynamic modes to ensure it can operate reliably in your SoC.
`EFLX Compiler Programming
`To program the EFLX array, use Synopsys® Synplify or another synthesis tool
`to generate an .edif file which is then input to the EFLX Compiler. You also
`provide an IO file which shows how the EFLX array is connected to the rest of
`your chip. EFLX Compiler packs, places, routes and iterates to generate
`optimum timing performance: if it is acceptable, EFLX Compiler generates the
`bitstream which when loaded into the EFLX array makes it execute your RTL.
`Our GUI makes array floorplan building and timing path analysis quick and easy. New features are in development as well.
`
`Assistance Throughout Your Chip Evaluation and Design
`From the start we will assign trained engineers to assist with you with evaluating our technology, presenting the details you need, assisting you with optimizing
`your RTL for embedded FPGA, and then working weekly through the whole physical design process. We will assist in DFT and chip test as well. Our
`engineers regularly visit China, Taiwan, Japan, Europe and Israel. Many of them are fluent in Mandarin and one in Hebrew.
`
`
`
`Case 5:18-cv-07581-LHK Document 31-9 Filed 03/04/19 Page 5 of 5
`
`EFLX embedded FPGA
`
`EFLX Embedded FPGA
`Evaluation Boards & Business
`EFLX200K T16FFC+ Evaluation Board
`Our most recent validation chip for TSMC16FFC+ EFLX4K implemented a full 7x7
`array, the EFLX200K is also designed as a customer evaluation chip and is now
`available on an evaluation board. The EFLX200K has 182K LUT4, 560 MACs,
`1.4Mbit attached SRAM, PLLs and PVT monitors. Using this customer can load
`even very large RTL to verify performance at full speed and at different tempera-
`tures and voltages; also customers can measure static and dynamic power. All
`future validation chips (up next is the TSMC28HPC+) will also be featured on
`evaluation boards for customer use.
`
`Business Model
`We have a competitive advantage: we can give you the exact eFPGA you need in any TSMC process node/variant, in the size you want, with the options you
`want and we can deliver it off the shelf already in multiple nodes and in ~6 months for any other TSMC node. And what we deliver is as dense and fast as full
`custom FPGA. Plus we can do this with less engineering-months than our competitors. We have done more ports in less time than our competitors.
`
`So we can afford to be flexible and very competitive on pricing. On ports to new TSMC nodes we don’t charge you for development: we price it the same as if
`it were already off the shelf.
`
`Our typical business model for a single design is: A) a license fee for any number and size/configuration of eFPGAs on your chip typically paid in milestones:
`contract, IP delivery, working silicon, B) a small royalty, pennies per chip, C) an annual seat license fee for the EFLX Compiler. We are open to other
`structures as well.
`Proven Management
`
`Our CEO was a GM at AMD since 1983 ending up as Senior VP of processors; then was founding CEO of Rambus joining as the 4th
`employee, achieving IPO 7 years later and running the company until it reached $2 Billion in Market Capitalization. Our Silicon Engineer-
`ing VP, Sales VP and Director of Solutions Architecture all have >20 years experience at (cid:31)rms like ARM, Intel, and Rambus. Our Software
`Director has >10 years experience at Synopsys and did his PhD thesis on FPGA Place & Route. Our Sr. Hardware Design Manager has >10
`years experience at Oracle designing in processes down to 7nm. Our Hardware Design Manager has done 8 FPGA/eFPGA designs.
`
`Well Financed
`
`We have raised ~$15 Million from Lux Ventures, Eclipse Capital & SVB.
`We have a strong cash balance. Our investors have deep pockets for additional capital as needed.
`We are committed to keep up with growing customer demand and to invest as necessary to do so.
`
`www.(cid:31)ex-logix.com
`Copyright © 2015-2018 Flex Logix Technologies, Inc. EFLX ,Flex Logix, XFLX, ArrayLinx & RAMLinx are Trademarks of Flex Logix Technologies, Inc.
`All other names mentioned herein are trademarks or registered trademarks of their respective owners
`5/2018TG
`
`