shakti/m_class/libre_riscv_chennai_2018.tex

   1 \documentclass[slidestop]{beamer}
   2 \usepackage{beamerthemesplit}
   3 \usepackage{graphics}
   4 \usepackage{pstricks}
   5
   6 \title{Commercial Libre-RISCV SoC}
   7 \author{Luke Kenneth Casson Leighton}
   8
   9
  10 \begin{document}
  11
  12 \frame{
  13    \begin{center}
  14     \huge{Designing a Commercial Libre RISC-V SoC}\\
  15     \vspace{32pt}
  16     \Large{Ethical Strategic Leveraging of the benefits}\\
  17     \Large{of Libre and Open SW/HW}\\
  18     \Large{for pure unadulterated Commercial gain}\\
  19     \vspace{24pt}
  20     \Large{Chennai 9th RISC-V Workshop}\\
  21     \vspace{16pt}
  22     \large{\today}
  23   \end{center}
  24 }
  25
  26
  27 \frame{\frametitle{Credits and Acknowledgements}
  28
  29  \begin{itemize}
  30    \item The Designers of RISC-V\vspace{15pt}
  31    \item The Shakti Group\vspace{15pt}
  32    \item Prof. G S Madhusudan\vspace{15pt}
  33    \item Neel Gala\vspace{15pt}
  34    \item Rishabh Jain\vspace{15pt}
  35   \end{itemize}
  36 }
  37
  38
  39 \frame{\frametitle{Why, How, What?}
  40
  41  \begin{itemize}
  42    \item Why? Because these days it's just not necessary to
  43          make [un]ethical compromises in order to make a profitable,
  44          desirable mass-volume product\\
  45          {\it (There's enough companies doing that: where it's got us??)}
  46    \item How? By leveraging the long-establised strategic cost and
  47          maintenance benefits of libre-licensed software (and
  48          HDL) and
  49          {\it making sure that the people who provide it are
  50          financially rewarded}.  Also by empowering diverse team
  51          collaboration
  52    \item What? A 2.5ghz RISC-V 64-bit SoC that has
  53                  a 3D Embedded GPU, 1080p Video decode, and interfaces
  54                  to make it attractive for use in tablets, netbooks, industrial
  55                  embedded and more.  22nm or less, under 400 pins, under USD \$4.\\
  56   {\it All sounds obvious... but is it practical and achievable?}
  57   \end{itemize}
  58 }
  59
  60
  61 \frame{\frametitle{Definitions}
  62
  63  \begin{itemize}
  64    \item {\bf Business}: the provision of a service and being
  65                  commensurately financially rewarded for doing so
  66    \item {\bf Spongeing}: the provision of a service and being
  67                   taken advantage of for doing so {\it (cf: Professor Yunus)}
  68         \item {\bf An ethical act}: an act that increases truth,
  69                   love, awareness or creativity for one or more people
  70                   (including yourself), {\it without} reducing those
  71                   same four qualities {\it for anyone}
  72         \item {\bf The Four Freedoms}: the rights and guarantees
  73                   associated with and embedded within GNU Licenses {\it (cf: FSF)}
  74   \end{itemize}
  75   {\it Is it possible to ethically do business and respect the
  76    Four Freedoms? That's where it gets interesting, as there are
  77    even cases where the Four Freedoms are unethical.  Note: google's
  78    former motto "don't be evil" is clearly (unintentionally) unethical}
  79 }
  80
  81
  82 \frame{\frametitle{Does what we want already exist?}
  83  \begin{center}
  84   \includegraphics[height=2.4in]{nolibresocs.jpg}\\
  85   {\bf Analysis of SoCs over the past 7+ years (answer: no)}
  86  \end{center}
  87 }
  88
  89
  90 \frame{\frametitle{What's the problem?}
  91
  92  \begin{itemize}
  93    \item {\bf iMX6}: Libre bootable, Vivante 3D GPU (libre etnaviv)
  94             but proprietary VPU (and a power-hungry Cortex A9)
  95         \item {\bf Allwinner SoCs}: mostly Libre bootable,
  96               VPU reverse engineered; GPU: MALI or PowerVR (i.e. proprietary)
  97         \item {\bf Rockchip SoCs}: good but using MALI or PowerVR.
  98         \item {\bf TI OMAP}: good but using PowerVR.  and expensive.
  99         \item {\bf Samsung}: good but using MALI.
 100         \item {\bf Ingenic jz4775}: GREAT! performance
 101                   sucks (1ghz MIPS32).
 102         \item {\bf Broadcom SoCs}: Cartelled. and boots from the GPU
 103   \end{itemize}
 104   {\it Basically there does not exist one single commercial SoC that
 105    provides full source code for all functions (CPU, GPU, VPU)
 106    with modern performance. Which is kinda bizarre if you think about it}
 107 }
 108
 109
 110 \frame{\frametitle{What would a good (Libre) boring, mundane SoC have?}
 111
 112  \begin{itemize}
 113    \item Cover a lot of different scenarios (embedded, tablets, industrial,
 114          netbooks, crypto-currency mining).
 115         \item Decent performance with high efficiency.  RISC-V: 40 \%
 116                   more efficient than ARM / Intel.  Shakti a good
 117               candidate: 2.5ghz and 120mW per core @ 22nm.
 118         \item 1080p video: y'all gotta watch cute kittens on youtube, right?
 119         \item 3D GPU: y'all gotta play Angri Burds, right? (or Minecraft)
 120         \item No spying back-door co-processors (to steal crypto-wallets)
 121         \item No Spectres, no Meltdowns.
 122   \end{itemize}
 123   {\it Basically quite boring and mundane.  No Monster Performance,
 124    no AI stuff, no special sauce. Just a plain-old SoC,
 125    40\% more power efficient than ARM/Intel,
 126    and not spying on end-users, that's all}
 127 }
 128
 129
 130 \frame{\frametitle{How on earth does an ethical Libre SoC make money???}
 131
 132  \begin{itemize}
 133    \item Simple answer: Mask Rights.
 134    \item Without Mask Rights: by having a desirable
 135          product, and packaging it for a customer (i.e. by being a middle-man
 136          a service is still being provided for which payment etc. etc.)
 137    \item Without a desirable product or customer(s): err... you don't.\\
 138              (cf: definition of Business)
 139    \item By not having high NREs (leveraging back-to-back deals,
 140              and helping others fulfil their needs and goals)
 141   \end{itemize}
 142   {\it Detachment from the goal also helps. If someone else makes this
 143    product then GREAT! I can go do something else}
 144
 145 }
 146
 147 \frame{\frametitle{Things wot are "off-limits"}
 148
 149  \begin{itemize}
 150    \item Customer entrapment (through proprietary software).\\
 151          Strong business case for not entrapping customers:\\
 152              https://tinyurl.com/most-productive-meeting-ever
 153    \item Funding, endorsing, supporting or empowering unethical
 154          Companies, Organisations, Cartels and Individuals.\\
 155              (cf: definition of an ethical act).
 156         \item Being totally inflexible / unrealistic.  Goals have
 157              to be met: it's no good being an idiot about that. e.g. if
 158              a Libre 3D GPU really can't be made, use Vivante GC800
 159              (with etnaviv).
 160   \end{itemize}
 161   {\it Still no real show-stoppers to making money (or product):
 162     it's just slightly harder, that's all.  Ultimately it's about
 163     confidence. }
 164 }
 165
 166
 167 \frame{\frametitle{Interfaces, Block Diagram, of the Libre-RISCV SoC}
 168  \begin{center}
 169   \includegraphics[height=2.1in]{../shakti_libre_riscv.jpg}\\
 170   {\bf Separate Power Domains for GPIO banks, Variable voltages
 171     required, low-power sleep states etc.  Quite involved}
 172  \end{center}
 173 }
 174
 175
 176 \frame{\frametitle{Hardware / Development Complexity Comparison}
 177
 178  \begin{itemize}
 179    \item {\bf Server}: relatively easy. PCIe, RapidIO, XAUI, SATA, GbE, 10GE,
 180                  DDR3/4 (or HMC) etc. etc. No multiplexing: all interfaces dedicated
 181                  and high-speed differential pairs.
 182    \item {\bf Desktop}: really just a variant of Server.
 183              Graphics is a PCIe Card (except if integrated).  Peripherals
 184              often done in dedicated external ICs ("Southbridge" concept)
 185    \item {\bf Embedded}: also pretty easy.  Really needs a pinmux.  Low clock
 186          rate, low power mode.  e.g. SiFive Freedom U310.
 187    \item {\bf Mobile}: HARD. Performance/Watt matters $=>$ variable core
 188                  voltage domains {\it per core}.  Number of pins matters (affects
 189                  yield and package cost).  Cost
 190                  matters.  Pinmux critical.
 191   \end{itemize}
 192   {\it Bottom line: Mobile-class processors are challenging!}
 193 }
 194
 195
 196 \frame{\frametitle{Proprietary vs Libre-licensed Interface HDL}
 197
 198  \begin{itemize}
 199    \item DDR3/4: challenging! \$1m for single-use, single instance.\\
 200          Symbiotic EDA: \$600k for PHY; CERN developed a Controller\\
 201          http://libre-riscv.org/shakti/m\_class/DDR/
 202    \item HyperRAM (JEDEC xSPI): lower risk than DDR3/4\\
 203            http://libre-riscv.org/shakti/m\_class/HyperRAM/
 204    \item RGMII: several available (saves \$50k)\\
 205            http://libre-riscv.org/shakti/m\_class/RGMII/
 206    \item UART, SPI, I2C, PWM, SD/MMC: all libre (except eMMC).
 207    \item Shakti Group has FlexBus, QuadSPI, SRAM, many more.
 208    \item RGB/TTL: R. Herveille (SSD2828, SN75LVDS83b, TFP410a)
 209   \end{itemize}
 210   {\it Basically there's no compelling reason to spend vast sums
 211    on proprietary HDL.  Sorry Cadence / Mentor / Synopsis / whoever}
 212 }
 213
 214
 215 \frame{\frametitle{Challenging Stuff [1] - Memory Interfaces}
 216
 217  \begin{itemize}
 218    \item DDR3/4 PHYs are analog and very high speed.
 219                    Impedance training.  Extreme timing tolerances on parallel buses.\\
 220                    No surprise they cost USD \$1m and above.
 221    \item Symbiotic EDA will do (Libre) PHY layout for USD \$300k,
 222              time to completion for chosen geometry: 8-12 months.
 223   \end{itemize}
 224    {\it Silicon-proven but still risky.  What are the alternatives?}
 225    \vspace{4pt}
 226  \begin{itemize}
 227    \item 133mhz 32-bit SDRAM (um...) maybe even FlexBus?
 228    \item HyperRAM (aka JEDEC xSPI) 8-bit SPI 166mhz or DDR-300.\\
 229            300mbyte/sec for only 13 wires, not bad!  (We'll take several)\\
 230            http://libre-riscv.org/shakti/m\_class/HyperRAM/
 231    \item HMC: insanely fast, very low power.  OpenHMC (LGPL)
 232            https://opencores.org/project/openhmc
 233   \end{itemize}
 234 }
 235
 236
 237 \frame{\frametitle{Challenging Stuff [2] - Video Decode Engine}
 238
 239  \begin{itemize}
 240    \item Richard Herveille's Video Core Blocks\\
 241             https://opencores.org/project/video\_systems
 242    \item Symbiotic EDA MP4 decoder in FPGA
 243    \item H.264 seems to have been done...\\
 244          https://github.com/adsc-hls/synthesizable\_h264
 245    \item Really needs SIMD (or better, not-SIMD)\\
 246          {http://libre-riscv.org/simple\_v\_extension/}
 247    \item Definitely needs xBitManip (parallelised by Simple-V)\\
 248          https://github.com/cliffordwolf/xbitmanip
 249   \end{itemize}
 250    {\it SIMD is insane. $O(N^6)$ opcode proliferation.  See\\
 251      https://www.sigarch.org/simd-instructions-considered-harmful/ \\
 252      (1): P-Ext designed for Audio. (2): Investigate RI5CY's SIMD
 253    }
 254 }
 255
 256
 257 \frame{\frametitle{Challenging Stuff [3] - 3D GPU.  Sigh.}
 258
 259  \begin{itemize}
 260    \item Actual requirements quite modest: 30MP/s 100MT/s 5GFLOPS
 261          but power/area is crucial ($2mm^2$ @ 40nm)
 262    \item Nyuzi, MIAOW, GPLGPU (Number Nine), OGP.
 263    \item Nyuzi based on Larrabee. Jeff Bush really helpful.
 264    \item MIAOW is an OpenCL engine.  GPLGPU is fixed-function
 265    \item Nyuzi lessons: Software-only rendering not enough.
 266              Getting through L1 cache takes most power. Fixed functions
 267              such as parallel FP-Quad to ARGB Pixel, and Z-Buffer
 268              needed.
 269    \item Fallback is GC800 (\$250k) {\it contact me if you can do better!}
 270   \end{itemize}
 271    {\it Jacob Bachmeyer's Cache-control proposal turns L1 Cache into
 272    scratchpad RAM.  RVV is just too heavy (sorry!), Simple-V much
 273    more light-weight and flexible.
 274    }
 275 }
 276
 277
 278 \frame{\frametitle{TODO}
 279
 280  \begin{itemize}
 281    \item TODO\vspace{8pt}
 282   \end{itemize}
 283 }
 284
 285
 286 \frame{\frametitle{Summary}
 287
 288  \begin{itemize}
 289    \item TODO
 290   \end{itemize}
 291 }
 292
 293
 294 \frame{
 295   \begin{center}
 296     {\Huge The end\vspace{20pt}\\
 297                    Thank you\vspace{20pt}\\
 298                    Questions?\vspace{20pt}
 299         }
 300   \end{center}
 301
 302   \begin{itemize}
 303         \item Discussion:
 304         \item http://libre-riscv.org/shakti/m\_class/
 305   \end{itemize}
 306 }
 307
 308
 309 \end{document}