Andes Technology Corp. · TWSE:6533
Andes Technology — Conference Presentation (April 2019)
23 slides from Andes Technology's conference presentation, April 2019
Slide 1: Introduction to Andes Superscalar Processors: N15(F)/D15(F)
Introduction to Andes Superscalar Processors: N15(F)/D15(F) Driving Innovations™ Vincent Chen Director 2018/7/31
Slide 2: N15(F)/D15(F) - Superscalar Processors
N15(F)/D15(F) - Superscalar Processors AndeStar™ V3 architecture Four AndesCore™ products N15 : Baseline ISA D15 : N15 + DSP extension ISA D15F : D15+ Floating point unit N15F : N15 + Floating point unit 6-stage dual-issue pipeline One instruction from ALU/Multiply/Divide/DSP/Branch/
Slide 3: Applications
Applications • Touch Screen • 3D printer • PLC • ADAS • IoT gateway • Industrial automation • NB-IoT • Smart home • Sensor hub RTOS/ Linux appliances • Home automation Bare metal • HMI control system • Wireless vacuum cleaner • Surveillance 32-bit dual-issue pipeline 2700 CoreMark @T40LP 4000
Slide 4: AndesCore: Product Lineup (V3 Arch.)
AndesCore™ : Product Lineup (V3 Arch.) Ultra Low Power Mainstream High Performance D15 E830 D15F D1088 N15 S801 N1337 N15F N1068 N968 Floating Point N820 N705 MMU(Linux) Static Branch Pred. I/D Cache BTB BTB FlashFetch AHB/AHB2/AXI Bus AHB-lite/APB Bus ECC 32-bit GPR, Multiplier, HW divider, Vector
Slide 5: AndesCore: Product Lineup (V3 Arch.)
AndesCore™ : Product Lineup (V3 Arch.) Bare Metal, RTOS-based Applications Linux-based Applications D15 E830 D15F D1088 N15 S801 N1337 N15F N1068 N968 Floating Point N820 N705 MMU(Linux) Static Branch Pred. I/D Cache BTB BTB FlashFetch AHB/AHB2/AXI Bus AHB-lite/APB Bus ECC 32-bit GPR, Multiplier, HW
Slide 6: Functional Blocks
Functional Blocks Pwr. mgmt. instruction 5-wire JTAG debug I/F Powerbrake & Trace I/F to Trace module QuickNap™ technology (EDM: Embedded Debug Module) 32 vector interrupts Standby Memory Mgmt. EDM & & VIC Trace I/F Unit (Linux ready) Single-cycle multiplier Memory protect Radix-4
Slide 7: StackSafe to Protect Stack Usage
StackSafe™ to Protect Stack Usage Recording mode: Record the maximum usage of stack pointer maximum 15 Stack Usage 10 5 0 time Protection mode: Allocate stack size and set its bound accordingly When stack pointer grows over the bound Exception ! 20 Generate an exception 15 Stack Usage 10
Slide 8: Power Management
Power Management PowerBrake to digitally adjust power (via stalling pipeline) QuickNap™: logic power-down and SRAM in retention Power maximum consumption performance minimum performance Standby QuickNap™ (logic power-down and SRAM in retention) Dormant SRAM power-down Shutdown Performance (frequen
Slide 9: AndeStar DSP ISA Extension (1/2)
AndeStar™ DSP ISA Extension (1/2) Features: > 130 instructions Saturation and Rounding Data types: fractional (Q31, Q15, and Q7) and integer (32b, 16b, 8b) 16-bit and 8-bit SIMD instructions 16-bit: +, -, x, min, max, abs, clip, compare, <<, >>, signed, unsigned 8-bit: +, -, min, max, abs
Slide 10: AndeStar DSP ISA Extension (2/2)
AndeStar™ DSP ISA Extension (2/2) SW support: Intrinsic functions for using instructions in C. Compiler generation of SIMD instructions based on vector data type. >200 DSP-ISA-optimized DSP libraries 110% performance boost with DSP ISA (average of integer functions) 10 Driving InnovationsTM
Slide 11: Andes DSP Library
Andes DSP Library Basic math: vector mathematics Fast math: sin, cos, atan, atan2, sqrt, etc. Complex math Statistics: max, min, RMS, etc. Filtering: IIR, FIR, LMS, etc Transforms: FFT, DCT/DCT4 Matrix functions PID controller, Clark and Park transforms MISC: copy/fill arrays, data type con
Slide 12: Floating Point Unit (FPU)
Floating Point Unit (FPU) IEEE 754 Compliant Single (32-bit) or Double (64-bit) precision configuration Register number configuration Single precision: 32 Double precision: 16 or 32 Support all rounding modes and exceptions Support Flush-To-Zero mode to speedup denormalized number proc
Slide 13: AMR-WB Performance
AMR-WB Performance N10 D10 D15 MAX 337.07 85.58 58.24 AMR-WB AVG 289.16 73.13 49.99 Encode MIN 188.50 51.47 33.20 MAX 75.27 25.21 15.25 AMR-WB AVG 66.61 21.96 13.54 Decode MIN 54.77 18.22 11.07 The benchmarking numbers are in MCPS. (Million Cycle Per Second) For all the test vectors, the maximum (MA
Slide 14: Support and Service
Support and Service 14 Driving InnovationsTM
Slide 15: Overview of Andes Technology Corporation
Overview of Andes Technology Corporation Andes Highlights • Founded in March 2005 in Hsinchu Science Park, Taiwan, ROC. • Core RD team from AMD, DEC, Intel, MIPS, nVidia, and Sun. • Just over 140 people; 80% are engineers. • EETimes' Silicon 60 Hot Startups to Watch (2012) • TSMC OIP Award “Partner
Slide 16: AndeSight: Professional IDE (1/2)
AndeSight™: Professional IDE (1/2) Eclipse-based Project Setup: Meta Linker Script Editor Flash ISP configured through GUI Debug Support: Script-Based RTOS Awareness Virtual Hosting Register Bitfield Viewing/Update Task Break-n-Display on Exceptions List Resource Usage 16 Driving I
Slide 17: AndeSight: Professional IDE (2/2)
AndeSight™: Professional IDE (2/2) Program Analysis Function Profiling Performance Meter Code Coverage Function Code Size (Static) Stack Size Debug/Analysis for Arduino Custom Plugin Intf 17 Driving InnovationsTM
Slide 18: AndeShape, AndeSoft IP Support
AndeShape™, AndeSoft™ IP Support AndeShape™ Platform IPs AE300: AXI fabric package for scalable SoC applications AE210P: generic platform IP for micro-controllers AXI Bus Components: AXI bus matrix, bus bridges AXI up/downsizer AHB Bus Components: AHB bus matrix, bus bridge, decoder
Slide 19: Andes IoT Stack
Andes IoT Stack ► Andes6 connectivity components ► Contiki RTOS for OS services ► An implementation of 6LoWPAN (IPv6 over 802.15.4) ► Commercial (e.g. InsideSecure) or open source TLS for security ► Connecting to the Clouds (Microsoft Azure, Acer BYOC) MQTT | CoAP TLS | DTLS TCP | UDP IPv6 Andes6 (6
Slide 20: Andes Ecosystem
Andes Ecosystem IP Design Service / Foundries SW and Development Environ. EDA/Others 20 Driving InnovationsTM
Slide 21: Andes Technical Support
Andes Technical Support Pre-Sales Post Sales • AndesCore™ Introduction • AndesCore™ Synthesis Report • Customizable Training Courses • Code Size Optimization • SoC Design Consulting • SoC Evaluation Consulting • E-service Support • A web-based supporting system • A systematic way to get Andes techni
Slide 22: Concluding Remarks (1/2)
Concluding Remarks (1/2) WHY Andes Patented processor architecture Leading performance efficiency Strong on cost performance Strong on power efficiency Intuitive standard SW development tools Comprehensive SW stacks Flexible platform solutions Elite third parties Responsive/experien
Slide 23: Concluding Remarks (2/2)
Concluding Remarks (2/2) WHY N15(F)/D15(F) Performance Extreme HIGH Computing Power 5.41 CoreMark/MHz; 3.36 DMIPS/MHz; 2700 CoreMark in TSMC 40LP process node; 4000 CoreMark in TSMC 28HPM process mode Low Latency and Fast Response Time Power Consumption Compact Core Optimized Memor
Andes Technology — Conference Presentation (April 2019) on 6ix