KYA chahiye hame: ek digital system ke behavioral description ko ek kaam karne wale physical chip mein badalna.
KYO split hai: yahan ek gehri engineering trade-off hai flexibility aur efficiency ke beech.
Agar tum apne design ke liye exact transistors banao (ASIC), tum sabse chhoti, sabse tez, sabse kam power wali chip paate ho — lekin tum ek bahut bada one-time cost chukate ho (masks, fabrication) aur ise bana lene ke baad badal nahi sakte.
Agar tum ek generic reprogrammable chip (FPGA) use karo, tum design ko seconds mein badal sakte ho aur up front almost kuch nahi chukate — lekin generic-ness ka tumhe area, speed, aur power mein nuksan hota hai.
Kis step par FPGA aur ASIC flows pehli baar diverge karti hain?
Technology mapping / physical implementation par — front end (spec, RTL, simulation, synthesis) shared hai.
FPGA par map kiya jaane wala atomic compute element kya hai?
Ek LUT (Look-Up Table), plus flip-flops, block RAM, DSP slices.
ASIC logic ko LUTs ki jagah kis cheez mein map karta hai?
Ek standard-cell library mein jo physically fabricate ki jaayegi.
FPGA back end ka final output kya hai?
Ek bitstream jo LUT contents aur routing switches configure karta hai.
ASIC back end ka final output kya hai?
GDSII masks jo foundry ko bheji jaati hain (tape-out).
Total cost model likho.
Ctotal(N)=CNRE+Ncunit.
Break-even volume formula derive karo.
FPGA cost = ASIC cost set karo, N ke liye solve karo: N∗=(CNREA−CNREF)/(cuF−cuA).
Post-fab bug ASIC ke liye catastrophic kyon hai lekin FPGA ke liye cheap?
ASIC ko mehenga mask re-spin chahiye; FPGA minutes mein sirf nayi bitstream reload karta hai.
CTS (Clock Tree Synthesis) kya karta hai, aur yeh ASIC-specific kyon hai?
Real wire delay equalize karne ke liye ek balanced clock distribution network banata hai; FPGA clock trees pre-built hote hain.
Bitstream aur GDSII mein fark?
Bitstream existing hardware configure karta hai; GDSII nayi geometry describe karta hai manufacture karne ke liye.
Zyattar FPGAs volatile kyon hote hain?
Woh bitstream SRAM mein store karte hain, jo power-off par config lose kar deta hai (flash se reload karna padta hai).
Break-even volume se neeche, kaun si technology jeeтti hai aur kyon?
FPGA — ASIC ki badi NRE, low volume par per-unit savings se recover nahi hoti.
Recall Feynman: ek 12-saal ke bacche ko samjhao
Socho tumne ek Lego robot invent kiya. Ek real banana ke tumhare paas do tarike hain.
Tarika 1 (FPGA): ek magic box khareed lo jo pehle se chhoti switches se bhari ho. Tum bas switches flip karo taaki box pretend kare ki woh tumhara robot hai. Agar galti ho, dobara flip karo — ek minute mein done. Box thoda chunky aur slow hai, aur unplug karne par sab bhool jaata hai.
Tarika 2 (ASIC): apna blueprint ek factory ko bhejo jo sand pighalaakar exact robot chip banaye. Woh chhoti hai, super fast hai, kam power peeti hai — lekin set up karna bahut mehenga hai aur agar blueprint mein typo thi, toh ise phenk do aur dobara pay karo.
Dono robot ki same drawing se shuru hote hain. Sirf drawing kaise real cheez banti hai mein fark hai.