Lesson 1/254%

MODULE 1 OF 25 12 MIN FPGA FUNDAMENTALS

Introduction to FPGA

Understand what a Field-Programmable Gate Array is, where it came from, and why it is one of the most important technologies in modern electronics.

Learning Objectives

Define FPGA and explain what "Field-Programmable" means
Describe the SRAM-based configuration model used by modern FPGAs
Trace the history of FPGA technology from 1984 to the present day
Identify the three fundamental programmable resources: Logic, Routing, and I/O
List at least six industries that rely on FPGA technology and explain why
Articulate why FPGAs are increasingly important as Moore's Law slows

What Is an FPGA?

A Field-Programmable Gate Array (FPGA) is a semiconductor device containing a vast array of configurable logic blocks, programmable interconnect, and flexible I/O cells — all of which can be wired together in any configuration the designer chooses. Unlike a fixed-function chip (a microprocessor, an Ethernet controller, a GPU), an FPGA ships from the factory as a blank slate. The manufacturer creates the silicon with all the potential connections in place; the user then loads a bitstream — a binary configuration file — that programs millions of tiny SRAM cells to define exactly which logic evaluates which signals and which wires connect to which.

The result is a device that literally becomes a custom digital circuit. Every clock cycle, the logic fabric you defined computes your algorithm entirely in hardware, with no instruction-fetch overhead, no operating-system interrupts, and no pipeline stalls caused by branch misprediction. If you design a 256-tap FIR filter, the FPGA fabric IS that filter — 256 multipliers running in parallel every single clock cycle.

Modern FPGAs from AMD Xilinx (Versal, UltraScale+), Intel (Agilex), and Lattice (ECP5, Nexus) contain millions of logic cells, hundreds of megabits of on-chip RAM, thousands of DSP slices, high-speed serial transceivers at 100+ Gbps, and in the latest devices, hardened AI inference engines. They are far more than "programmable logic" — they are complete heterogeneous computing platforms in a single package.

Definition FPGA stands for Field-Programmable Gate Array. "Field" means programmable after manufacturing — in the field, by the customer. "Gate Array" refers to the array of logic gates on the silicon that can be interconnected in any combination.

Engineering Tip Think of an FPGA as a blank whiteboard made of silicon. You draw your own logic circuit on it using RTL code (VHDL or Verilog), and the FPGA synthesizes and becomes that circuit. Change the drawing, reload the bitstream — the chip becomes something completely different.

A Brief History of FPGAs

The FPGA was invented by Ross Freeman and Bernard Vonderschmitt, who co-founded Xilinx in 1984. Freeman's key insight was that PLAs (Programmable Logic Arrays) could be extended into a much larger, more general architecture using SRAM cells to configure routing. The first commercial FPGA, the XC2064, appeared in 1985 and contained just 64 Configurable Logic Blocks (CLBs) and 58 I/O blocks — barely enough to implement a simple finite state machine.

FPGA Evolution Timeline

1984

Xilinx founded by Ross Freeman and Bernard Vonderschmitt in San Jose, California. The FPGA concept is born.

1985

XC2064 — first commercial FPGA. 64 CLBs, 58 I/O pads, 2-micron CMOS process. Limited to simple glue logic but proved the concept.

1990s

CPLD vs FPGA era. Altera introduces CPLDs; Xilinx XC4000 series adds dedicated carry chains and RAM. FPGAs begin displacing 7400-series logic boards.

2000

Virtex-II Pro introduces hardened PowerPC processor cores embedded in FPGA fabric. The era of heterogeneous computing begins.

2006

Virtex-5 brings 65nm process, 6-input LUTs, and high-speed serial transceivers (3.75 Gbps). FPGAs enter serious telecom and defense applications.

2010

Zynq-7000 SoC — ARM Cortex-A9 + FPGA fabric on one die. Enables embedded Linux alongside custom hardware acceleration.

2018

UltraScale+ at 16nm FinFET. HBM (High Bandwidth Memory) integrated on package, 100G+ transceivers. Data center era begins.

2022

Versal ACAP — not just an FPGA, but an Adaptive Compute Acceleration Platform with hardened AI Engines (400 TOPS), scalar processing, and DSP arrays.

FPGA Evolution at a Glance

graph LR A["1984\nXilinx founded"] --> B["1985\nXC2064\n64 CLBs"] B --> C["1990s\nCPLD era\nXC4000"] C --> D["2000s\nEmbedded CPUs\nMicroBlaze/PPC"] D --> E["2006\nVirtex-5\nHigh-speed Serial"] E --> F["2010\nZynq SoC\nARM+FPGA"] F --> G["2018\nUltraScale+\nHBM Integration"] G --> H["2022\nVersal ACAP\nAI Engine 400 TOPS"] style A fill:#1e3a5f,stroke:#22d3ee,color:#e2e8f0 style H fill:#1e3a5f,stroke:#34d399,color:#e2e8f0

FPGA Resource Growth Over Generations

The Three Fundamental Programmable Resources

Every FPGA, regardless of vendor or generation, is built around three foundational programmable resource categories. Understanding these is the foundation for everything else in this course.

🧠

Logic — LUTs and Flip-Flops

Lookup Tables (LUTs) implement any boolean function of N inputs by storing a truth table in SRAM. Flip-Flops register outputs for pipelined designs. Together they form the Configurable Logic Block (CLB) — the workhorse of every FPGA.

🔗

Routing — Programmable Interconnect

A hierarchy of programmable switch matrices and wire segments that can be configured to carry any signal to any destination on the chip. Routing accounts for 50–70% of FPGA silicon area and is often the critical path timing bottleneck.

📡

I/O — Configurable IO Buffers

Programmable IO blocks (IOBs) implement hundreds of different voltage standards (LVCMOS, LVDS, SSTL) and drive strengths. High-speed serial transceivers (GTX, GTH, GTY) handle 10–112 Gbps per lane for PCIe, Ethernet, and optical interfaces.

Interview Question Q: What makes an FPGA different from a microcontroller?
A: An FPGA implements custom hardware logic in parallel — it IS the circuit. A microcontroller executes instructions sequentially in software on a fixed CPU core. The FPGA fabric has no instruction fetch cycle; every logic element evaluates every clock cycle simultaneously.

Common Mistake Beginners often think FPGAs "run programs" like a CPU. They don't — they ARE the hardware. Your RTL code (VHDL/Verilog) describes a physical circuit, not a sequence of instructions. The synthesis tool converts your RTL into gates and flip-flops; the place-and-route tool assigns them to physical FPGA resources.

Industries That Rely on FPGAs

FPGAs appear in a remarkable diversity of industries, often hidden inside products you interact with daily. Their unique combination of high performance, reconfigurability, and longevity makes them irreplaceable in many applications.

Industry	Key Applications	Why FPGA Wins	Example Device
Aerospace & Defense	Radar DSP, EW, satellite comms, radiation-hardened control	Rad-tolerant designs (RH FPGA families), long product life (20+ years), deterministic timing	Xilinx Virtex-5QV (space-grade)
Telecommunications	5G base stations (fronthaul/midhaul), OTN switching, packet processing	Flexible protocol implementation, 400G+ line rate processing, low latency PHY	Xilinx UltraScale+ VU9P
Data Center	SmartNICs, network acceleration, ML inference, database offload	Programmable network pipelines, 100G/400G wire-rate processing, HBM bandwidth	Xilinx Alveo U55C, Versal AI Core
Automotive	ADAS sensor fusion, LiDAR processing, radar signal chain, ISO 26262 control	Functional safety (lockstep), real-time processing, sensor interface flexibility	Xilinx Zynq UltraScale+ MPSoC
Medical	Ultrasound beamforming, CT reconstruction, implantable device control, MRI	Ultra-low latency, custom ADC/DAC interfaces, FDA validation longevity	Xilinx Artix-7, Kintex-7
Broadcast	4K/8K video processing, SMPTE routing, real-time encode/decode	Multiple simultaneous video pipelines, standards agility (SDI, HDMI, NDI)	Xilinx UltraScale+ VU27P
Industrial	Motor control, machine vision, PLC I/O, real-time EtherCAT	Custom real-time control loops, microsecond response, custom fieldbus protocols	Xilinx Zynq-7000, Artix-7
Financial (HFT)	Order management, market data feed processing, risk analytics	Sub-microsecond order-to-send latency, direct hardware I/O (QSFP), zero OS jitter	Xilinx Alveo U50, UltraScale+ VU13P

Why FPGAs Matter More Than Ever

For decades, Moore's Law provided "free" performance — every two years, transistors halved in size and you got 2x the compute for the same power. That era is effectively over. At 5nm and below, each new node costs billions more to develop, delivers diminishing density gains, and faces fundamental physical limits. Designing a custom ASIC at cutting-edge nodes now requires NRE (Non-Recurring Engineering) costs of $50M–$500M.

FPGAs fill the gap perfectly. They are manufactured on the latest process nodes (TSMC 7nm for UltraScale+, TSMC 7nm for Versal) by AMD, meaning you get the transistor density and speed of leading silicon without paying NRE. You get hardware-level performance and energy efficiency, with the ability to change what the hardware does via a software bitstream update — at any time, anywhere in the world.

The rise of AI/ML workloads is the most recent driver. Neural network inference requires enormous arithmetic throughput but must run at the edge with tight power and latency budgets — a regime where FPGAs excel over both CPUs and GPUs. AMD's Versal AI Core with its AI Engine array demonstrates this convergence: programmable logic meets hardened neural network compute on the same die.

Best Practice When evaluating a design for FPGA implementation, ask three questions: (1) Does the algorithm require parallel execution? (2) Is latency below 1 microsecond critical? (3) Does the design need a custom electrical interface not available on standard processors? If "yes" to any, FPGA is worth a serious evaluation.

Knowledge Check

1. What does the acronym FPGA stand for?

Fully Programmable Gate Array
Field-Programmable Gate Array
Fixed-Program Generic Architecture
Flexible Programmable Grid Array

Correct! FPGA = Field-Programmable Gate Array. "Field" indicates it can be programmed after manufacturing, in the field by the end user.

2. Who is credited with inventing the commercial FPGA?

Gordon Moore (Intel)
Jack Kilby (Texas Instruments)
Ross Freeman (Xilinx)
Andy Grove (Intel)

Correct! Ross Freeman co-founded Xilinx in 1984 and invented the FPGA concept. He later received the Computer Entrepreneur Award from the IEEE.

3. What type of memory is used to store the configuration of most modern FPGAs?

Flash NOR memory
EEPROM cells
SRAM (Static RAM)
Anti-fuse elements

Correct! Most commercial FPGAs (Xilinx, Intel) use SRAM-based configuration cells. This means the configuration is volatile — you must reload the bitstream from external flash on every power cycle. Anti-fuse FPGAs exist for space/defense but are one-time programmable.

4. FPGAs execute instructions sequentially, just like a CPU. True or False?

True — FPGAs use a fetch-decode-execute pipeline
False — FPGAs evaluate all logic in parallel every clock cycle

Correct — False! FPGAs don't execute instructions at all. Every logic block (LUT, flip-flop) in the design evaluates every clock cycle simultaneously. There is no instruction pointer, no fetch cycle, and no sequential bottleneck.

5. Which of the following is NOT one of the three fundamental programmable resources in an FPGA?

Programmable Logic (LUTs and Flip-Flops)
Programmable Routing (Interconnect)
Cache Memory (L1/L2 cache)
Programmable I/O (IOBs)

Correct! Cache memory (L1/L2) is a CPU architecture concept. FPGAs do not have cache. The three fundamental programmable resources are Logic (LUTs+FFs), Routing (programmable interconnect), and I/O (configurable IO buffers).

Practical Exercise — FPGA in the Wild

Research Challenge: Find FPGAs in Everyday Products

FPGAs are far more common than most engineers realize. Your challenge: find 3 products you interact with regularly that contain an FPGA. For each product, identify the FPGA vendor, the device family, and the specific function the FPGA performs.

Start with networking equipment. Enterprise WiFi access points (Cisco, Aruba, Juniper) frequently contain Xilinx or Lattice FPGAs for packet classification and hardware offload. Search "[product model] FCC ID" then look up the FCC test report which often lists IC part numbers.

Look at test & measurement. Digital oscilloscopes (Keysight, Tektronix, Rigol) almost universally use FPGAs for real-time waveform acquisition and triggering. Check the product teardown on EEVblog or similar forums.

Investigate broadcast or pro-audio gear. Professional video routers (Blackmagic Design, Ross Video) and high-end audio interfaces use FPGAs for custom signal processing and interface standards (SDI, ADAT).

Create a table with columns: Product | Brand/Model | FPGA Vendor | Device Family | Function Performed. Aim for at least 3 entries. Bonus: find an FPGA in a medical device or automotive product.

Hint Anandtech teardowns, iFixit, and TechInsights are great resources for identifying silicon inside consumer and enterprise products. Search "[product] teardown PCB" on Google Images for IC part numbers.

Lesson Summary

Key Takeaways

An FPGA is a silicon chip whose logic connections are defined by the user via a configuration bitstream, not by the manufacturer
Modern FPGAs use SRAM cells — volatile, requiring bitstream reload on power-up from external non-volatile memory
Ross Freeman (Xilinx) invented the commercial FPGA in 1984–1985; the XC2064 had 64 CLBs
The three fundamental resources: Logic (LUTs + Flip-Flops), Routing (programmable interconnect), I/O (configurable IO buffers)
FPGAs are used across aerospace, telecom, data center, automotive, medical, broadcast, industrial, and financial industries
As Moore's Law slows and ASIC NRE costs skyrocket, FPGAs provide hardware performance at software flexibility — becoming more critical every year
FPGAs do NOT execute instructions — they ARE the circuit. All logic evaluates in parallel every clock cycle

← Course Overview Why FPGA? →