A-Level Computer Science

The three number systems 数制 you must use:

• denary 十进制 (decimal, base 10) — uses digits 0–9. Place values are powers of ten.
• binary 二进制 (base 2) — uses 0 and 1. Place values are powers of two. Every byte 字节 is 8 bits 位.
• hexadecimal 十六进制 (base 16) — uses 0–9 then A–F for 10–15. Each hex digit 数位 stands for exactly 4 bits.

Conversions

Denary → binary: keep dividing by 2 and record the remainders, read bottom-up. Or subtract the largest place value 位值 (power of 2) that fits.

Example: $558_{10}$: $558 = 512 + 32 + 8 + 4 + 2 = 2^{9} + 2^{5} + 2^{3} + 2^{2} + 2^{1}$. In 12 bits: 0010 0010 1110.

Binary → hex: group the bits into nibbles 半字节 (4 bits) from the right and convert each. 0010 0010 1110 → 2 2 E → 22E.

Hex → binary: replace each hex digit with its 4-bit pattern. Hex → denary: multiply each digit by its place value. 22E $= 2 \times 256 + 2 \times 16 + 14 = 558$.

Worked example. Convert denary 200 to 8-bit binary, then to hexadecimal.

$200 = 128 + 64 + 8$, so the binary is 11001000. In nibbles, 1100 1000 $= 12$ and $8$, i.e. $\text{C}$ and $8$, so the hexadecimal is C8.

Binary vs decimal prefixes

Two prefix families look similar but differ — decimal (powers of 10) and binary (powers of 2):

So a tebibyte (TiB) is slightly more than a terabyte (TB). A "1 TB" drive holds $10^{12}$ bytes, but an operating system that reports in TiB shows a smaller number.

Binary addition

Add column by column from the right, carrying as in denary:

Overflow 溢出 happens when the result needs more bits than the register 寄存器 can hold — the carry-out of the leftmost column is the overflow bit.

Binary subtraction

The usual way is two's complement 补码 addition: to do $A - B$, form the two's complement of $B$ (invert every bit and add 1), then add, and discard any final carry-out.

To subtract $00011110$ from $01100100$ (unsigned 8-bit):

• two's complement of $00011110$: invert → $11100001$, add 1 → $11100010$.
• add to $01100100$: result $1\,01000110$ (9 bits) — discard the leading 1 → $01000110 = 70_{10}$. Check: $100 - 30 = 70$. ✓

Two's complement signed integers

In an $n$-bit two's-complement number:

• the most significant bit 最高有效位 (MSB) is the sign bit 符号位: 0 = positive, 1 = negative.
• to read a negative number: invert every bit, add 1, then negate.

So $11100010$ is negative; invert → $00011101$, add 1 → $00011110 = 30$, so it is $-30$. This is a signed integer 有符号整数 (unlike an unsigned 无符号 one). The range for $n$ bits is $-2^{n-1}$ to $+2^{n-1} - 1$; for 8 bits, $-128$ ($10000000$) to $+127$ ($01111111$).

Worked example. What denary value does the 8-bit two's-complement number $10110100$ represent?

The MSB is 1, so it is negative. Invert → $01001011$, add 1 → $01001100 = 76$, so the value is $-76$. Check with place values: $-128 + 32 + 16 + 4 = -76$.

Overflow in signed arithmetic happens when the true result falls outside this range — spotted when the sign bit flips wrongly (two positives giving a negative, or two negatives giving a positive).

One's complement

Before two's complement, an older scheme called one's complement 反码 represented a negative number by simply inverting every bit of the positive — there is no "add 1" step.

• $+30 = 00011110$, so in one's complement $-30 = 11100001$ (just the inverse).
• Drawback: it has two zeros — $00000000$ ($+0$) and $11111111$ ($-0$) — which wastes a bit pattern and makes arithmetic awkward.

Two's complement (invert and add 1) removes the negative zero: it has a single zero and lets addition and subtraction use the same circuit. That is why modern computers store signed integers in two's complement, not one's complement.

In BCD 二进码十进数, each denary digit is written as its own 4-bit pattern. The number $93$ is 1001 0011 in BCD — not binary 93 ($01011101$). Each nibble uses only 0–9; patterns $1010$–$1111$ are invalid.

BCD reading: 0010 0111 0101 → 2, 7, 5 → 275.

Use: calculators, digital clocks, and devices that show denary digits — each digit drives a 7-segment display 七段显示器. Currency code often uses BCD to avoid the rounding errors of converting fractions like 0.1 to binary.

Hex is a compact way to write binary (1 hex digit = 4 bits):

• memory addresses 内存地址 in low-level programming — 0x7FFE.
• colour values in HTML/CSS — #FF8800.
• MAC addresses — AC:DE:48:00:11:22.

Hex does not change the stored data — it just makes binary easier for humans.

Computers store text as numbers; each character has a numeric code point 码点 set by a character set 字符集.

ASCII

• ASCII uses 7 bits — 128 code points. Basic Latin letters, digits, punctuation, and control codes.
• Extended ASCII uses 8 bits — 256 code points; the lower 128 match ASCII, the upper 128 vary by region.

Unicode

• Unicode is a universal character set covering almost every script, plus symbols and emoji.
• common encodings 编码: UTF-8 (1–4 bytes, ASCII-compatible), UTF-16 (2 or 4 bytes), UTF-32 (fixed 4 bytes).

Why Unicode beats ASCII

• it represents far more characters (every script, emoji); ASCII covers only basic English.
• files are portable with no code-page confusion, and allow multilingual text in one document.
• trade-off: Unicode files are usually larger for English-only text.

A bitmap 位图 image (also called a bitmapped image) stores the colour of every pixel 像素 in a grid. At the start of the file a file header 文件头 records the image's metadata — its width, height and colour depth — so software knows how to read the pixel data that follows.

• image resolution 图像分辨率: the bitmap's own size, width × height in pixels (e.g. 1920 × 1080).
• screen resolution 屏幕分辨率: the width × height the display can show. If an image's resolution is larger than the screen it is scaled down to fit; a low-resolution image looks blocky when stretched onto a higher-resolution screen.
• colour depth 颜色深度 (bit depth 位深度): bits per pixel. 1 bit → black/white; 8 bits → 256 colours; 24 bits → 16.7 million ("true colour").

File size

Divide by 8 for bytes, by 1024 for KiB, etc. Example: a $3000 \times 2000$ image at 24 bpp is $3000 \times 2000 \times 24 = 1.44 \times 10^{8}$ bits $\approx 17.2\ \text{MiB}$.

Changing settings

• lower resolution → smaller file, less detail (looks blocky when enlarged).
• lower colour depth → smaller file, but smooth shades show banding.
• higher of either → larger file, better quality.

A vector graphic 矢量图形 stores the instructions to draw the image as a drawing list 绘图列表 — an ordered list of drawing objects 绘图对象 (geometric primitives 图元: lines, curves, polygons, circles). Each drawing object has properties 属性 such as colour, fill, line width and position (coordinates). To show it, the program renders 渲染 the drawing list at any resolution needed.

Bitmap vs vector

Vector advantage: it scales without losing quality — a vector logo stays sharp at any size, while a bitmap blurs when enlarged. Vector disadvantage: it cannot describe arbitrary pixel detail (photographs).

A continuous wave of analogue data 模拟数据 (the sound) is converted into digital data 数字数据 by sampling 采样:

• sampling rate 采样率 — samples per second (Hz). CD quality is $44.1\ \text{kHz}$.
• sampling resolution 采样分辨率 (bit depth) — bits per sample's amplitude 振幅. CD quality is 16 bits.

File size

A 10-second stereo CD clip: $44100 \times 16 \times 10 \times 2 = 14\,112\,000$ bits $\approx 1.68\ \text{MiB}$.

Changing settings

• higher sampling rate → captures higher pitches, larger file.
• higher sample resolution → finer amplitude steps, less quantisation 量化 noise, larger file.
• lower of either → smaller file, clear quality loss.

(The sampling rate must be at least twice the highest frequency you want to keep.)

Compression 压缩 reduces file size, saving storage and transmission bandwidth 带宽. Two kinds:

• lossless 无损 — the original data is recovered exactly (text, programs, ZIP/PNG).
• lossy 有损 — some detail is dropped for much smaller files (JPEG, MP3, video).

When to use which

• lossless for documents, source code, medical images — anything needing exact data.
• lossy for streaming media. Real-time video streaming uses lossy compression because it must send huge amounts of data in real time over limited bandwidth; lossless would not shrink it enough. Raw HD video is gigabytes per minute, so without compression the picture would keep freezing.

Lossless methods

• run-length encoding 行程编码 (RLE): store "the next $n$ values are $x$" instead of repeating $x$. Great for flat areas; useless for noisy data.
• dictionary methods 字典编码 (ZIP, PNG): replace repeated byte sequences with a short reference. Good for text and code.
• Huffman coding 霍夫曼编码: give short codes to common symbols and long codes to rare ones, bringing the average code length near the data's entropy 熵.

Lossy methods

• images (JPEG): drop fine detail and colour differences the eye barely sees.
• sound (MP3, AAC): drop pitches we hear less well, and quiet sounds hidden by louder ones.
• video combines spatial 空间 compression (within each frame, like JPEG) with temporal 时间 compression (most frames store only the differences from the previous frame).

• Show working for base conversions: denary → binary by place values, binary → hexadecimal in nibbles (groups of 4 bits).
• For two's complement the MSB is negative; to negate, invert and add 1; watch for overflow when the sign bit flips wrongly.
• Distinguish bitmap (pixels; file size $=$ width $\times$ height $\times$ colour depth) from vector (drawing commands; scales without loss).
• Sound file size depends on sample rate $\times$ bit depth $\times$ time — more of each means better quality but a bigger file.
• Compare lossless vs lossy compression and give a use for each.

A network 网络 is a set of computing devices connected so they can communicate and share resources. Benefits:

• sharing resources (printers, file servers, internet) — cheaper than equipping each computer.
• sharing data — many users access the same files.
• central management — install software, manage users and back up once on a server.
• communication — email, video calls, messaging.
• remote access — work from anywhere.

A local area network 局域网 (LAN) covers a small area — a home, office or school, usually owned by the organisation, with high data rates and low latency 延迟.

A wide area network 广域网 (WAN) covers a large area — a city, country, or the world (the internet is the largest WAN). It uses telecom-company infrastructure — often the Public Switched Telephone Network 公共交换电话网 (PSTN), leased lines or fibre — with lower data rates and higher latency. A WAN connects LANs together.

Client-server

• powerful machines act as servers 服务器, providing services (files, web pages, email).
• other machines are clients 客户端 that request services.
• central and easy to manage, but the server is a single point of failure unless backed up.

Peer-to-peer (P2P)

• all machines are equal peers; each can be both client and server (peer-to-peer 对等网络).
• resources are spread across the peers — no central server. Robust to one failure, but harder to keep secure and consistent.

A thin client 瘦客户端 does little processing locally and relies on a powerful server (web terminals, remote desktops). A thick client 胖客户端 has strong local processing and storage and runs full applications itself (a normal desktop PC).

The topology 拓扑 is how the nodes and links are arranged.

• bus 总线 — all devices on one shared cable. Cheap; the whole LAN fails if the bus fails; performance drops as more devices share the bandwidth 带宽.
• star 星形 — every device connects to a central switch. One device failing does not affect others; the switch failing brings all down. Most common today.
• mesh 网状 — every device links directly to others, with many paths. Very fault-tolerant 容错 (traffic reroutes) but needs lots of cabling.
• hybrid — a mix (a star in each office, mesh links between offices).

Cloud computing 云计算 delivers computing services (servers, storage, software) over the internet, hosted by a third party. Benefits: scalability 可扩展性 (pay for what you need), lower cost, access from anywhere, and reliable redundant data centres. Drawbacks: needs internet, your data is held by a third party, and possible vendor lock-in.

• wired (Ethernet 以太网 over twisted-pair 双绞线 or fibre-optic 光纤): higher speed, lower latency, fewer errors, more secure.
• wireless (Wi-Fi, Bluetooth, cellular): no cables, devices can move, but slower, prone to interference and eavesdropping.

For the same generation, wired wins on speed and reliability; wireless wins on convenience.

• network interface card 网络接口卡 (NIC) — lets a device send and receive on the network; has a unique MAC address MAC地址 (a 48-bit hardware address). A wireless device uses a wireless network interface card 无线网络接口卡 (WNIC).
• switch 交换机 — forwards Ethernet frames only to the port for the destination MAC address.
• hub 集线器 — a simpler device that copies traffic to all ports (now obsolete).
• wireless access point 无线接入点 (WAP) — lets wireless clients join a wired LAN.
• cabling — twisted-pair for short runs; fibre-optic for longer, faster runs.

A router 路由器 connects different networks and forwards data between them — usually at the boundary of a LAN and the internet. It does:

• forwarding — reads each packet 数据包's destination IP address IP地址 and sends it out the right port, using a routing table 路由表.
• network address translation 网络地址转换 (NAT) — lets many private LAN addresses share one public IP.
• DHCP 动态主机配置协议 — hands out private IP addresses to LAN devices.
• firewall 防火墙 — blocks unwanted incoming traffic.

Ethernet is the main wired LAN technology. On shared media a collision 冲突 can happen when two devices send at once. The protocol is CSMA/CD 载波侦听多路访问/冲突检测 (Carrier Sense Multiple Access with Collision Detection):

1. carrier sense — listen before sending; wait if the cable is busy.
2. multiple access — many devices share the medium.
3. collision detection — keep listening while sending; a clash is a collision.
4. on a collision, both stop, send a brief "jam" signal, then wait a random backoff time before retrying.

Modern switched Ethernet uses full-duplex 全双工 point-to-point links, so collisions no longer happen.

Bit streaming 流式传输 sends multimedia as a continuous stream that the receiver plays as it arrives, instead of downloading the whole file first.

• real-time (live): captured and streamed as it happens (live sport, video calls). You cannot rewind; low latency is vital.
• on-demand: pre-recorded on a server (YouTube, Netflix). You can pause and rewind; the server can buffer 缓冲 ahead.

Real-time streaming works as a short pipeline:

1. capture and sample the source (a camera or microphone).
2. encode it, using compression 压缩 to shrink the data.
3. send it across the network as packets.
4. the receiver buffers a little, then plays it live — dropping any packet that arrives late, because a live stream cannot wait for it.

Lossy 有损 compression is used here: moving pictures hide small losses, and the stream must be small enough to fit the bandwidth.

The internet 互联网 is a global network of networks using a common protocol 协议 suite (TCP/IP). The World Wide Web 万维网 (WWW) is a service that runs over it: hyperlinked documents identified by URLs, viewed in browsers via HTTP/HTTPS. Email and file transfer are other internet services that are not part of the WWW.

An IP address uniquely identifies a device.

• IPv4 — 32-bit, four denary numbers 0–255 (192.168.1.10); about $4.3 \times 10^{9}$ addresses (now exhausted).
• IPv6 — 128-bit, eight groups of four hex digits; about $3.4 \times 10^{38}$ addresses.

Subnetting

A network can be split into subnets 子网. The IP address splits into a network part and a host part, given by a subnet mask 子网掩码 (e.g. 255.255.255.0 = first 24 bits are network). Subnetting improves management, cuts broadcast traffic, and improves security.

Public vs private addresses

• private addresses are used within a LAN and are not routable on the internet (e.g. 192.168.0.0/16).
• a public IP address is globally unique and routable, assigned by an ISP 互联网服务提供商.

Devices behind NAT with private addresses are not directly reachable from the internet, giving some protection.

Static vs dynamic

• a static IP address is fixed; used for servers that must be found at a known address.
• a dynamic IP address is assigned by DHCP and may change; easier for client devices and uses a limited address pool efficiently.

Worked example. A host has IP address 192.168.10.130 with subnet mask 255.255.255.192. Which network is it on, and is 192.168.10.200 on the same one? The mask's last octet, 192, is 11000000 in binary, so the first 26 bits are the network part and the last 6 bits address the host. That makes the subnets step in blocks of $256 - 192 = 64$: .0, .64, .128, .192. The address 130 falls in the block starting at .128, so the host is on network 192.168.10.128/26, whose usable hosts run .129 to .190 (.191 is the broadcast address). 200 falls in the next block (.192), so it is on a different subnet and traffic between the two must pass through a router. Get the block size from the mask first ($256$ minus the mask octet) - guessing from the first three octets is what makes these go wrong.

A URL 统一资源定位符 (Uniform Resource Locator) locates a resource on the WWW:

https://www.example.com/about/contact.html
protocol     domain name        path

• protocol: http, https, etc.
• domain name 域名: a readable server address.
• path: the resource on that server.

The Domain Name System 域名系统 (DNS, also called the Domain Name Service) is a distributed set of servers that turns domain names into IP addresses. When you type a URL, the browser asks a DNS resolver for the IP, which queries DNS servers (root → top-level → authoritative) until it finds it; the browser then connects to that IP and requests the path. DNS saves humans from memorising IP addresses and lets a site change server without changing its name.

• Distinguish LAN vs WAN and client-server vs peer-to-peer by who stores and controls the resources.
• Match each topology (bus, star, mesh) to its advantages and drawbacks (cost, reliability, collisions).
• Know the job of each device: a switch directs within a LAN by MAC address, a router routes between networks by IP.
• Explain bit streaming and why buffering is needed (data arrives at a different rate from playback).
• Distinguish IPv4 vs IPv6 and public vs private addresses; DNS turns a URL into an IP address.

A general-purpose computer has four building blocks:

• input devices 输入设备 — get data in (keyboard, mouse, microphone, scanner, sensors).
• output devices 输出设备 — give results out (monitor, speakers, printer, actuators).
• primary memory 主存储器 — fast memory the processor 处理器 (CPU) reaches directly (RAM and ROM). Holds the running program and its data.
• secondary storage 辅助存储器 — slower, larger, keeps programs and data when not in use (hard disk, SSD, optical disc, USB stick).

An embedded system 嵌入式系统 is a computer built into another device to do one fixed job (washing machine, microwave, car engine unit, thermostat).

• benefits: optimised for one task (low power, small, cheap); reliable; fast to start; cheap in volume.
• drawbacks: limited to its one task; hard to update (its firmware 固件 may need special tools); often not repairable; sometimes weak security.

Laser printer

A laser printer 激光打印机 scans the page image onto a charged photosensitive drum 感光鼓. Toner 墨粉 sticks to the charged areas, transfers to the paper, and is melted on by a fuser. Fast, sharp, high-volume.

3D printer

A 3D printer 3D打印机 builds an object layer by layer: FDM melts plastic filament through a nozzle; stereolithography cures liquid resin with a UV laser. Used for prototypes and custom medical parts.

Microphone and speakers

A microphone 麦克风 turns sound into an electrical signal (a diaphragm vibrates, changing capacitor 电容器 charge or coil position); the signal is digitised by an analogue-to-digital converter 模数转换器 (ADC). A speaker does the reverse — a varying signal drives a coil in a magnetic field, moving a cone to make sound.

Magnetic hard disk (HDD)

A hard disk 硬盘 stores data on spinning platters coated with magnetic material. Each platter has tracks 磁道 divided into sectors 扇区. A read/write head 读写头 floats just above and magnetises tiny regions (write) or senses them (read). Cheap per gigabyte, but slower than SSDs and has moving parts.

Solid-state (flash) memory

A solid-state drive 固态硬盘 stores data as charge in transistors 晶体管, with no moving parts. Faster random access than HDDs, tougher, lower power, but dearer per gigabyte; each cell wears out after many writes.

Optical disc

A laser detects reflections from tiny pits on an optical disc 光盘 (CD, DVD, Blu-ray). The drive is an optical disc reader/writer: writing uses a stronger laser to change the surface's reflectivity.

Touchscreen

A touchscreen 触摸屏 senses contact. Resistive 电阻式: two conductive layers pressed together; works with anything but is less accurate. Capacitive 电容式: a finger disturbs a charge field; accurate, multi-touch, used in phones.

Virtual reality headset

A virtual reality 虚拟现实 (VR) headset has two small displays (one per eye) and motion sensors (accelerometer 加速度计, gyroscope 陀螺仪) that track head movement so the scene shifts as you look around.

A buffer 缓冲 is memory that holds data temporarily while it moves between devices of different speeds. Example: the CPU writes a document to a printer buffer quickly, then is free to do other work while the printer prints from the buffer at its own pace. Buffers stop the fast device waiting for the slow one (also used in streaming, the keyboard, and disk access).

• RAM 随机存取存储器 (Random Access Memory) — volatile 易失性 (loses data without power). Holds the OS, running programs and their data; read and written constantly.
• ROM 只读存储器 (Read-Only Memory) — non-volatile 非易失性 (keeps data without power). Usually written once; holds firmware needed at start-up (the BIOS / boot loader).

ROM starts the system; RAM then holds the active work.

• SRAM 静态RAM (Static RAM) stores each bit in a flip-flop 触发器 of several transistors. Fast, but expensive and not dense. Used for CPU cache 高速缓存.
• DRAM 动态RAM (Dynamic RAM) stores each bit as charge on a tiny capacitor. Cheaper and denser but slower, and must be refreshed 刷新 (rewritten) thousands of times a second. Used for main memory.

Use SRAM for small fast memory (cache); DRAM for large main memory.

ROM variants you can program after manufacture:

• PROM (Programmable ROM) — written once (fuses burned by a programmer); cannot be changed.
• EPROM (Erasable Programmable ROM) — erased by strong UV light through a window, then rewritten (whole chip at once).
• EEPROM (Electrically Erasable Programmable ROM) — erased and rewritten electrically, a byte at a time, in circuit. Flash memory is a derivative optimised for block erase.

Both read sensors; the difference is what they do next.

• monitoring 监控 — collects and reports data but takes no action (a weather station logging readings).
• control system 控制系统 — uses sensor data to decide and act through actuators, usually in a feedback loop (a thermostat turning a boiler on/off).

Sensors and actuators

A sensor 传感器 turns a physical quantity into a signal: temperature (a thermistor 热敏电阻 or thermocouple), pressure (strain gauge), infra-red, sound. Analogue signals need an ADC first. An actuator 执行器 does the reverse — turns a signal into an action (a motor, valve, heater, buzzer).

Feedback

In a control system the actuator changes the environment, which the sensors then re-measure — a feedback 反馈 loop. Without feedback the system cannot correct itself or know when to stop (a thermostat with no temperature feedback would heat forever).

A logic gate 逻辑门 is a small circuit that does one Boolean 布尔 operation. Inputs and outputs are 0 (false, low) or 1 (true, high). Know the symbol, function and truth table 真值表 for each gate.

NOT (inverter)

AND — output 1 only if all inputs are 1

OR — output 1 if at least one input is 1

NAND (NOT AND) — output 0 only when all inputs are 1

NOR (NOT OR) — output 1 only when all inputs are 0

XOR (Exclusive OR, also called EOR) — output 1 if the inputs are different

A logic circuit 逻辑电路 is a network of gates that carries out a Boolean expression. You should be able to move between a problem statement, a logic expression, a truth table, and a circuit diagram.

From expression to circuit

Draw one gate per operator and wire them up. For $X = (A \text{ AND } B) \text{ OR } (\text{NOT } C)$: a NOT gate on $C$, an AND gate on $A$ and $B$, then an OR gate on the two results.

From circuit to expression

Work forwards from the inputs, labelling each gate's output, until you reach the final output.

From circuit to truth table

For $n$ inputs there are $2^{n}$ rows. List every input combination; for each, work out the internal gates then the output.

From truth table to expression (sum of products)

For each row that outputs 1, write an AND of the inputs (with NOT on any input that is 0 in that row); OR these together. Example: a table that is 1 only on $(A=0,B=1)$ and $(A=1,B=0)$ gives $\overline{A}B + A\overline{B}$, which is $A \text{ XOR } B$.

From a problem statement

Turn the English into a Boolean expression first: "A and B" → A AND B; "A or B or both" → A OR B; "exactly one of A and B" → A XOR B; "neither A nor B" → A NOR B; "not both" → A NAND B.

Worked example. A machine's alarm $X$ sounds when the guard is open ($A=1$) and either the motor is running ($B=1$) or the temperature is high ($C=1$). Write the Boolean expression, and give the rows where $X=1$. Turn the English into logic one clause at a time: "either B or C" is $B + C$, and "A and that" is $X = A\cdot(B + C)$. For the rows, $X=1$ needs $A=1$ and at least one of $B$, $C$ equal to 1 - so $(A,B,C) = (1,0,1)$, $(1,1,0)$ and $(1,1,1)$, three rows out of eight. Notice $A=0$ can never sound the alarm, whatever $B$ and $C$ do. Bracket the OR before ANDing it: $X = A\cdot B + C$ is a different circuit altogether, one that would sound the alarm on a high temperature even with the guard closed.

• Distinguish RAM (volatile, read/write) from ROM (non-volatile, holds the bootstrap); SRAM (cache, faster) from DRAM (main memory, needs refreshing).
• For a logic circuit, build the Boolean expression gate by gate, then a truth table covering every input combination.
• Learn the symbol, expression and truth table for each gate (AND, OR, NOT, NAND, NOR, XOR).
• Explain a buffer (a temporary store bridging two different speeds) and the role of an interrupt.

The Von Neumann architecture 冯·诺依曼体系结构 underlies almost every general-purpose computer:

• a single memory — the Immediate Access Store 立即存取存储器 (IAS) — holds both program instructions and data (the stored program 存储程序 concept).
• a processor 处理器 (CPU) fetches instructions from memory and runs them one at a time.
• instructions run in order unless a branch changes the flow.

The stored-program idea is what makes a computer flexible: change the program and you change what it does, with no rewiring.

All of these parts sit inside one small chip. The diagram later in this section shows how they connect; the photo below shows the real thing.

Arithmetic and Logic Unit (ALU)

The ALU 算术逻辑单元 does the arithmetic (add, subtract, …) and logic (AND, OR, comparisons). It takes operands from registers 寄存器 and puts results back in a register.

Control Unit (CU)

The control unit 控制单元 decodes each instruction and sends the control signals to carry it out — opening data paths, telling the ALU what to do, and controlling memory reads and writes.

System clock

The clock sends a steady stream of pulses that keep the CPU in step. Each instruction takes a fixed number of cycles, and the clock speed 时钟频率 (e.g. 3.8 GHz) is one factor in performance.

Registers

Registers are tiny, very fast stores inside the CPU. The special purpose registers 专用寄存器 each have a fixed job in the cycle:

• Program Counter 程序计数器 (PC) — the address of the next instruction.
• Memory Address Register 内存地址寄存器 (MAR) — the address being read or written.
• Memory Data Register 内存数据寄存器 (MDR) — the data going to or from memory.
• Current Instruction Register 当前指令寄存器 (CIR) — the instruction being decoded.
• Accumulator 累加器 (ACC) — the value the ALU is working on.
• Status Register 状态寄存器 — holds flags 标志 (carry, zero, negative, overflow) used by branches.
• Index Register 变址寄存器 — an offset used in indexed addressing.

General-purpose registers 通用寄存器 are used by the programmer for temporary values during a calculation. Movements of data between registers and memory are written in register transfer 寄存器传送 notation — e.g. MAR ← [PC] ("copy the contents of PC into MAR").

Three internal buses 总线 (sets of parallel wires) connect the parts:

• address bus 地址总线 — carries the memory address. One-way (CPU → memory).
• data bus 数据总线 — carries the data. Two-way.
• control bus 控制总线 — carries control signals (read, write, interrupt). Two-way.

An $n$-bit address bus can reach $2^{n}$ memory locations. The data-bus width sets how many bits move per access (often the word size).

• clock speed — more cycles per second.
• number of cores 核心 — a multi-core CPU runs several threads at once.
• word size 字长 — a 64-bit CPU handles 64-bit chunks per cycle and can address far more memory than a 32-bit one.
• amount of RAM 随机存取存储器 — more RAM holds more of the working set; too little forces the OS to page 页 to disk.
• cache memory 高速缓存 size — more cache cuts average memory access time.
• secondary storage 辅助存储器 type — an SSD loads programs far faster than an HDD.
• bus width and speed — wider/faster buses move data more quickly.

Match the specs to the workload: a quad-core beats a dual-core on parallel work, but higher per-core speed wins on single-threaded work.

A port 端口 is a physical socket for connecting a peripheral 外围设备:

• USB (Universal Serial Bus) — general-purpose (keyboards, drives, phones).
• HDMI (High Definition Multimedia Interface) — digital video and audio to a screen.
• VGA (Video Graphics Array) — older analogue video output to a monitor.
• Ethernet (RJ-45) — wired LAN. Audio jacks — headphones/microphone.

Different ports use different signals, so an HDMI cable will not fit a USB socket. USB-C is unusual in carrying video, data and power.

The CPU repeats the fetch-execute cycle 取指-执行周期, one run per machine instruction.

Fetch

1. the PC's address is copied to the MAR.
2. the PC is incremented to point to the next instruction.
3. a read signal goes over the control bus.
4. memory puts the instruction on the data bus.
5. it is copied into the MDR, then into the CIR.

Decode

The CU decodes the instruction in the CIR — what operation, and which operands or addresses.

Execute

The CU carries it out: arithmetic/logic goes to the ALU (result to the ACC); a load/store moves data between memory and a register; a branch changes the PC. Then the cycle repeats.

An interrupt 中断 is a signal that pauses the normal cycle so the CPU can handle an urgent event (a key press, a packet arriving, a hardware fault, division by zero, the OS timer).

Handling one:

1. finish the current instruction.
2. save the state (PC and registers).
3. load the address of the interrupt service routine 中断服务程序 (ISR) into the PC and run it.
4. the ISR handles the event.
5. restore the saved state and carry on.

Interrupts let the system respond promptly without the CPU constantly checking devices, and are how the OS multitasks.

The CPU actually runs machine code 机器码 — bit patterns, specific to one architecture. Assembly language 汇编语言 is a readable form, with one instruction per machine instruction, written using mnemonics 助记符 like LDD, ADD, JMP. An assembler 汇编器 translates it to machine code.

Two-pass assembler

A two-pass assembler reads the source twice:

• pass 1 builds a symbol table 符号表: each time a label 标签 (like LOOP:) appears, record its address; no code yet.
• pass 2 generates code: translate each instruction, and when one refers to a label (like JMP LOOP), look up its address in the symbol table.

Two passes handle forward references 前向引用 (a jump to a label defined later).

Example instruction set

Cambridge uses a small generic set: data movement (LDD, LDM, LDI, LDX, STO, MOV), arithmetic (ADD, SUB, INC, DEC), logic/bit (AND, OR, XOR, LSL, LSR), compare and branch (CMP, JMP, JPE, JPN), I/O (IN, OUT), and END. The exact mnemonics are given in the paper's reference table.

The addressing mode 寻址方式 (the modes of addressing) says how the CPU finds the operand:

• immediate addressing 立即寻址 — the operand is the value in the instruction. LDM #10 loads 10.
• direct addressing 直接寻址 — the instruction holds an address; the operand is the value there. LDD 200.
• indirect addressing 间接寻址 — the instruction holds an address that holds another address, which is the data. LDI 200.
• indexed addressing 变址寻址 — effective address is address + index register; used for arrays. LDX 100 with IR = 5 reads address 105.

(Relative addressing 相对寻址 gives the address as an offset from the PC — used for jumps.)

Worked example. Memory holds: location 200 = 250, location 250 = 99, location 105 = 7. The index register holds 5. What is in the accumulator after each of LDM #200, LDD 200, LDI 200 and LDX 100? Follow how far each mode has to look. LDM #200 is immediate - the operand is the number written in the instruction, so the accumulator holds 200. LDD 200 is direct - go to location 200 and take what is there: 250. LDI 200 is indirect - location 200 holds 250, which is another address, so go on to location 250: 99. LDX 100 is indexed - add the index register to the address, $100 + 5 = 105$, and read location 105: 7. Count the hops to keep them apart: immediate 0, direct 1, indirect 2, indexed 1 (once the index has been added).

To trace it: make a table with columns for the PC, ACC, index register, each variable and any flags. Step through the instructions, updating the table after each; follow branches when they change the PC; stop at END. A common pattern is a loop over an array using indexed addressing.

A logical shift 逻辑移位 moves all the bits left or right by some places, filling new positions with 0.

• left shift by 1 (LSL #1) — bits move left, a 0 enters on the right; for an unsigned number this is × 2.
• right shift by 1 (LSR #1) — bits move right, a 0 enters on the left; for an unsigned number this is integer ÷ 2.

Shifting by $n$ places multiplies or divides by $2^{n}$. Example: 00001011 (11) LSL #1 → 00010110 (22).

An arithmetic right shift keeps the sign bit so a negative signed number stays negative. A cyclic shift 循环移位 (rotate) feeds the bit that drops off one end back in at the other end, so no bits are lost.

Bit manipulation for monitoring/control

Embedded devices often use one bit 位 of a register per signal (e.g. bit $n$ = LED $n$). Using a mask 掩码 — bit masking — you can:

• set bit $n$: R = R OR a mask with bit $n$ set.
• clear bit $n$: R = R AND a mask with bit $n$ clear and the rest set.
• toggle bit $n$: R = R XOR a mask with bit $n$ set.
• test bit $n$: R AND the mask, then check if the result is non-zero.

Bit manipulation is fast, uses little memory, and lets one byte hold up to 8 on/off states.

• Learn the fetch-execute cycle in register-transfer terms (PC, MAR, MDR, CIR, ACC) and what increments the PC.
• Name each register's job; the address bus is one-way, the data bus is two-way.
• Distinguish the addressing modes (immediate, direct, indirect, indexed) — a frequent question.
• Explain how clock speed, number of cores, cache size and word length affect performance.
• For a binary shift, state whether it is logical or arithmetic; a left shift multiplies by 2, a right shift divides by 2.

Why a computer needs an OS

Hardware on its own can only fetch and run instructions — it knows nothing about files, programs, networks or users. The operating system 操作系统 (OS) is the software layer that:

• manages the hardware (processor 处理器, memory, I/O, storage) for the running programs.
• provides services (file system, network, user accounts) through a clear interface, so programs need not talk to the hardware directly.
• provides a user interface (command line, GUI, touch).
• lets several programs share the hardware safely — each gets fair CPU time and is kept out of the others' memory.

Without an OS, every program would need its own drivers, and only one program could safely run at a time.

Key management tasks

• process management 进程管理 — load programs, schedule them on the CPU, switch between them, and kill misbehaving ones. (A running program is a process 进程.)
• memory management 内存管理 — give memory to processes, keep them apart, and use virtual memory 虚拟内存 / paging 分页 so the working set can exceed physical RAM 随机存取存储器.
• file management — organise files and folders on secondary storage 辅助存储器, control permissions, prevent write corruption.
• device management (hardware management) — handle I/O and peripherals through device drivers 设备驱动, buffer data, manage interrupts, and give a uniform interface.
• security management — accounts, permissions, firewall, encryption.
• user interface and networking.

Utility software

Most OSes include utility programs 实用程序 to maintain the system:

• disk formatter — a file / disk management tool that prepares a disk (sets up its file system); related tools copy, move and delete files.
• defragmentation software (disk defragmenter 碎片整理) — moves the pieces of fragmented files together on a hard disk to cut seek time (not useful on SSDs).
• disk contents analysis / disk repair software — check disk integrity and fix file-system errors and bad sectors.
• back-up software (backup 备份) — copies user data elsewhere so it can be recovered.
• virus checker (antivirus 杀毒软件) — scans for malware and quarantines threats.
• firewall 防火墙 — filters network traffic by rules.
• file compression (compression 压缩) / archiving; system monitor; updates.

Bundling these with the OS saves the user installing each one.

A program library 程序库 is pre-written code (subroutines 子程序, classes, modules) that programs reuse instead of writing it themselves — e.g. a maths library, a network library, a graphics library.

Benefits: saves time (off-the-shelf code), reliable (well-tested, widely used), and standardised (consistent behaviour).

• a static library 静态库 is copied into the executable at compile time (stands alone, but larger and needs rebuilding to update).
• a dynamic library 动态库 (DLL, Dynamic Link Library; .so) is loaded at run time (smaller executables, shared by many programs, updated once for all).

You write source code; the computer runs machine code 机器码. A translator 翻译器 converts between them.

Assembler

An assembler 汇编器 translates assembly language 汇编语言 into machine code, one instruction per instruction. Used for low-level code (embedded systems, drivers).

Compiler

A compiler 编译器 translates a high-level program into machine code once, before it runs.

• it reports all errors at compile time; once clean, it produces a stand-alone executable 可执行文件 that runs without the compiler installed and can be run many times.
• generally faster at run time (no translation while running), but tied to one CPU/OS — recompile for each platform.

Interpreter

An interpreter 解释器 translates and runs a high-level program one line at a time, producing no executable.

• it reports an error when it reaches that line, then stops; you can fix it and continue — good for development.
• the interpreter must be installed to run the program; generally slower (each run re-translates), but easy to port across platforms.

Choosing between them

Hybrid: Java

Java is compiled into bytecode 字节码 (a platform-independent intermediate form), which a virtual machine 虚拟机 (the JVM) then interprets — or uses just-in-time compilation 即时编译 to turn hot parts into native code. So errors are caught early, the bytecode runs anywhere with a JVM ("write once, run anywhere"), and long-running programs reach near-native speed. C# and Python use similar designs.

Worked example. Java source is compiled to bytecode, which a JVM then interprets. Why use both, instead of compiling straight to machine code? A compiler produces machine code for one processor and operating system, so a program compiled on one machine will not run on another. Java's compiler instead targets a virtual machine, so the bytecode it produces is identical everywhere; each platform then supplies its own JVM to interpret that bytecode into its own native instructions. One compiled file therefore runs anywhere a JVM exists - "write once, run anywhere". The price is speed: interpreting bytecode is slower than running native code, which is why a real JVM also uses JIT compilation to turn frequently-run bytecode into native code while the program runs. Name both sides - the marks are for portability bought at the cost of speed.

An integrated development environment 集成开发环境 (IDE) brings the tools to write, test and debug code into one application:

• source code editor with syntax highlighting 语法高亮 (keywords, strings, comments in different colours), auto-indent and bracket matching.
• auto-complete 自动补全 — suggests names and shows function parameters as you type.
• translator integration — compile/run with one keystroke; errors shown inline.
• debugger 调试器 — set breakpoints 断点 to pause, step through line by line, and inspect variables.
• version control 版本控制 integration (git), project management, a help system, refactoring 重构 tools (safe renaming), and unit test 单元测试 integration.

An IDE speeds development by putting writing → running → debugging → fixing behind one interface. Common IDEs: Visual Studio, PyCharm, Eclipse, VS Code.

• List the OS's jobs (memory, process, file, device and security management) — "manages resources" alone is too vague.
• Compare compiler vs interpreter vs assembler: what each translates and when errors are reported.
• Explain what an IDE provides (editor, debugger, error diagnostics, auto-complete).

These sound alike but mean different things:

• security 安全 — protecting data from unauthorised 未授权 access, change or destruction.
• privacy 隐私 — an individual's right to control who sees their personal data, with consent and a clear purpose.
• integrity 完整性 — the data being accurate and complete — not corrupted or accidentally changed.

A file can be secure (only the right people can open it) but lack integrity (a typo corrupted it); or accurate but not private (anyone can read it). All three are needed.

Two things to protect: the data itself (keep it confidential, intact and available) and the computer system (a compromised system can attack others, steal credentials, or be held to ransom).

Threats fall into three groups.

1. Malware 恶意软件 (malicious software) — harmful programs:

• virus 病毒 — self-copying code that attaches to other programs and spreads when they run.
• worm 蠕虫 — self-copying code that spreads over networks 网络 with no user action.
• Trojan horse 木马 — looks useful but hides malicious code.
• spyware 间谍软件 — secretly collects information (keystrokes, passwords).
• ransomware 勒索软件 — encrypts your files and demands payment.
• adware 广告软件 — pushes unwanted adverts.

2. Tricking people (social attacks):

• phishing 网络钓鱼 — fake emails/sites that trick users into giving credentials.
• pharming 域名欺骗 — redirects a user to a fake site even when they type the correct address.
• social engineering 社会工程 — tricking people into giving up information.

3. Attacks on the network:

• hacking 黑客入侵 by hackers 黑客 — unauthorised access, often via weak passwords or software flaws.
• denial of service 拒绝服务 (DoS/DDoS) — floods a server so real users cannot reach it.
• eavesdropping 窃听 — capturing data in transit (a risk on open Wi-Fi).
• man-in-the-middle 中间人攻击 — an attacker secretly relays or alters messages between two parties.

Measures protect both the security of data (against loss, theft or corruption) and the security of the computer system (its hardware, software and network).

A standalone PC

• a strong password; antivirus kept up to date; prompt software updates; backup 备份 to separate media; full-disk encryption 加密; a locked screen.

A networked PC

All the above, plus a firewall 防火墙, per-user permissions (admin rights only for admins), central management of user accounts 用户账户, and audit logs 审计日志 (who logged in, what they touched).

Across the internet

• VPN 虚拟专用网 — encrypts traffic between the user and the corporate gateway.
• HTTPS / TLS — encrypt web traffic.
• digital signatures 数字签名 — prove who sent a message and that it was not altered in transit.
• intrusion detection — watches traffic for known attack patterns.

• interception in transit → encrypt the data (HTTPS, VPN). Intercepted ciphertext is useless without the key.
• unauthorised access → strong authentication 身份验证 (long passwords; two-factor authentication 双因素认证 with a phone code or key); user authorisation 授权; lock-out after failed logins.
• malware → anti-virus software and anti-spyware 反间谍软件 with real-time scanning; patching; avoid untrusted downloads.
• phishing → user training; email filtering; check the URL before entering credentials.
• internal threats → the least-privilege 最小权限 principle (give each user only what they need); auditing.
• DDoS → rate limiting and traffic filtering.

• encryption — turn plaintext 明文 into ciphertext 密文 with a key. Symmetric encryption 对称加密 (AES) uses one shared key; asymmetric encryption 非对称加密 (RSA) uses a public key 公钥 and a private key 私钥. Protects data at rest and in transit.
• access control 访问控制 — file permissions (read/write/execute) and access rights 访问权限, enforced by the OS.
• authentication — authentication techniques verify the user: something you know (password), have (token, phone), or are (biometrics 生物识别 — fingerprint, face, iris); strongest combined.
• backups — keep copies (some off-site) so loss or corruption is recoverable.
• physical security — locked server rooms, cable locks.

Data has integrity when it is accurate and complete. Two techniques: data validation (catch bad data before storing) and data verification (confirm data was entered or transferred correctly).

Validation — does the data make sense?

Validation 验证 checks data against sensible rules, automatically:

• range check — within limits (a month is 1–12).
• limit check — on the correct side of a single limit (e.g. age ≥ 18).
• existence check — the referenced item exists (e.g. a product code is in the table).
• length check — the right number of characters.
• type / character check — the right kind of data (a phone field allows only digits).
• format check — matches a pattern (an email must contain @).
• presence check — required fields are not empty.
• check digit 校验位 — an extra digit computed from the others (ISBN, card numbers) that spots transcription errors.
• lookup check and consistency check (e.g. delivery date ≥ order date).

Validation catches data that is wrongly formatted, but not data that is the right format yet factually wrong ("Bob" for "Bib").

Verification — was the data entered or transferred correctly?

Verification 核对 checks the data was not changed in moving from one place to another.

During entry: double entry (type it twice and compare, as for a new password) or visual check.

During transfer (bits can flip):

• parity check 奇偶校验 — an extra bit makes the number of 1s even (even parity) or odd. The receiver re-counts. Catches single-bit errors.
• checksum 校验和 — the sender sends a summary value of the data; the receiver recomputes it and compares.
• cyclic redundancy check 循环冗余校验 (CRC) — a stronger checksum using polynomial division, catching many more error types.

A parity block check 奇偶块校验 goes further and locates the error. Arrange the bytes in a grid: give each byte a row parity bit, then compute one extra parity byte whose bits are the column parity of the bytes above. A single flipped bit now fails one row and one column – their intersection pinpoints exactly which bit changed, so it can even be corrected.

Verification only proves what arrived matches what was sent — not that the data is correct, and not against deliberate tampering. Validation asks "is this sensible?"; verification asks "was this copied correctly?" — use both.

Worked example. A user types their date of birth as 31/02/2009, and types their email address twice. Which check catches which error, and what is the difference? Validation asks "is this data sensible?" - the computer tests it against a rule, and a format or range check rejects 31/02/2009 because February never has 31 days. Verification asks "was this data entered correctly?" - typing the email twice is double entry, and comparing the two copies catches a typing slip. The limit is what makes this a favourite question: validation can never tell you the data is right, only that it is possible - 01/02/2009 passes every validation rule even if the user was actually born on a different day. Say what each check can and cannot catch.

• Keep the three ideas separate: security (keeping data safe), privacy (who may see it), integrity (keeping it correct).
• Match each threat (malware, hacking, phishing, interception) to a measure (firewall, encryption, authentication, access rights).
• Encryption protects confidentiality, not integrity — use a checksum, parity or check digit for integrity.
• Distinguish a virus, worm and Trojan and how each spreads.

A computing professional is someone whose work — software, systems, networks, data — affects other people. Because the work is technical, others often cannot judge whether it was done well or honestly. So the profession follows shared ethics 伦理 (principles for good behaviour).

Why ethics matters

• trust — users and employers trust professionals to act in their interest. Without that trust, software loses credibility.
• impact — software runs medical devices, banking, vehicles. Careless or dishonest work can hurt people.

Professional bodies (BCS, ACM, IEEE) publish codes of ethics for members.

Typical principles

• public interest first — protect the safety and welfare of those affected.
• honesty and competence — be honest about your skills; don't claim expertise you lack.
• confidentiality 保密性 — protect clients' and employers' private information.
• avoid conflicts of interest 利益冲突 — don't take work where your interest clashes with the client's.
• keep your skills current; respect intellectual property 知识产权 and privacy 隐私; treat colleagues fairly.

Acting ethically vs unethically

Acting ethically protects users, strengthens reputation, reduces legal risk, and builds trust. Acting unethically (skipping testing, hiding bugs, misusing data) can harm real users, lead to dismissal or legal action, damage reputation, and erode trust in technology generally.

When you face a borderline decision: identify whose interests are affected, check the code of ethics and the law, weigh the consequences, ask a trusted senior, and choose the option that protects users above short-term convenience.

Worked example. Your team's new AI hiring tool sorts CVs ten times faster, but you notice it rejects more older applicants. Shipping it pleases your manager, but it treats one group unfairly. The ethical choice is to hold it back until the bias is fixed — public interest and fairness come before short-term convenience.

Copyright 版权 is the legal right of the creator of an original work to control how it is copied, distributed, modified and performed. It applies automatically (no registration) to source code, software, documents, images, audio and video.

Without copyright, anyone could copy software freely, the developer would not be paid, and plagiarism would be legal. With copyright, developers can earn from their work (encouraging more software), users know who made it, and re-use happens on the developer's terms through licensing. Copyright lasts a long time (often 70 years after the creator's death). General ideas and algorithms are not covered by copyright but may be covered by a patent 专利.

A software licence 软件许可证 is a contract granting permission to use software on the owner's terms; choosing and applying one is called software licencing.

Commercial (proprietary)

• commercial software is sold: you buy a licence; the software is used only within its terms.
• the source code is not given (a proprietary 专有 product); you cannot modify or redistribute it.
• examples: Microsoft Office, Adobe Photoshop, most games.

Used when the developer wants revenue per user and to keep control of the code.

Open-source

• the source code is public; users can read, modify and redistribute it (open-source 开源).
• permissive licences (MIT, BSD) allow almost any use; copyleft 著佐权 licences (GPL) require that modified versions are released under the same licence ("share-alike").
• the Free Software Foundation (FSF) and the Open Source Initiative (OSI) promote and approve open-source licences.
• examples: Linux, Python, Apache.

Used when the developer wants the software widely used and improved by the community.

Freeware and shareware

• freeware 免费软件 — free of charge, no source code, may be redistributed but not modified (Acrobat Reader, WhatsApp).
• shareware 共享软件 — free for a trial period, then you pay to keep using it; no source code.

To justify a licence choice, link it to the developer's goal (revenue, reach, community), the user's needs (cost, customising), and the use case.

Artificial intelligence 人工智能 builds systems that do tasks once thought to need human intelligence — recognising speech and images, translating, playing games, driving.

Most modern AI uses machine learning 机器学习 — algorithms that improve at a task by learning patterns from large amounts of data, instead of being programmed step by step. Deep learning 深度学习, using neural networks 神经网络 with many layers, is the leading approach today.

Everyday examples

AI tasks split into two kinds — understanding input, and producing output or decisions.

Understanding input:

• speech recognition 语音识别 — spoken words to text (voice assistants).
• image recognition 图像识别 — finding objects, faces or text in images.

Producing output or decisions:

• machine translation 机器翻译 — automatic translation between languages.
• recommendation systems 推荐系统 — suggesting products, videos or music.
• autonomous vehicles 自动驾驶汽车 and robots.

A common exam scenario: a program reads a label with a camera, translates it, and reads it aloud — using optical character recognition 光学字符识别 to find the words, machine translation to convert them, and text-to-speech 文本转语音 for the audio.

Benefits

• accessibility — speech/image AI helps users with impairments; translation helps non-native speakers.
• productivity — automating repetitive tasks frees people for creative work.
• decision support — AI spots patterns in huge datasets (medical diagnosis, fraud detection).
• always available, and personalised to each user.

Concerns

• bias 偏见 — unfair patterns in the training data become unfair AI decisions (hiring, lending).
• job displacement — AI may replace some roles.
• privacy — training often uses large amounts of personal data.
• transparency — large models are "black boxes", hard to explain.
• accountability — when AI is wrong, who is responsible: developer, user, or operator?
• misuse — deepfakes, misinformation, surveillance.

Professionals must understand the limits of the AI they build, inform users, and reduce harm.

• Answer ethics questions against a professional code of conduct (public interest, competence, honesty), not personal opinion.
• Distinguish copyright (protects the expression) from a patent (protects an invention).
• Compare software licences: proprietary, open-source, freeware, shareware and FOSS.

Before databases, programs stored data in flat files 平面文件 — usually one file per program. This is fine for small data but breaks down at scale.

Limitations

• data redundancy 数据冗余 — the same data (a customer's address) is held in several files.
• data inconsistency 数据不一致 — redundant copies updated separately get out of sync.
• data dependence — programs are tied to the file format; change the format and every program must be rewritten.
• hard to enforce integrity 完整性, hard to share safely, weak querying, and weak per-field security.

A relational database 关系数据库 fixes these by storing data in tables managed by one piece of software (the DBMS) that all programs use.

• table 表 (relation) — a grid of rows and columns; one table per type of entity 实体 (e.g. CUSTOMER).
• record 记录 (row, also called a tuple 元组) — one row; one instance of the entity.
• field 字段 (column, also called an attribute 属性) — one column; one piece of information about each record.
• primary key 主键 — a field (or fields) that uniquely identifies each record; never null or duplicated.
• foreign key 外键 — a field whose value matches the primary key of another table, linking the two.
• composite key 复合键 — a primary key made of two or more fields together.
• candidate key 候选键 — any field(s) that could be the primary key.
• secondary key 次键 — a non-primary field that is indexed for fast searching.
• indexing 索引 — building an index on a field so look-ups and joins run faster.
• referential integrity 参照完整性 — every foreign-key value must match an existing primary key (no orphan records).

A table is written in shorthand with the primary key underlined and foreign keys noted:

CUSTOMER(CustomerID, Name, Phone)
ORDER(OrderID, CustomerID, OrderDate)   -- CustomerID is FK → CUSTOMER