The opening questions of June 2025 Paper 1 (9618/12) hit three of the most exam-heavy AS topics: the Von Neumann model and interrupts, number system conversions, and an AI application with software licensing. These reward precision — vague answers bleed marks here — so let's pin down exactly what's expected.
📄 The original paper and mark scheme are on the Cambridge International past papers page. Questions below are paraphrased and explained in my own words, with worked examples I've written myself.
Von Neumann architecture & interrupts
The first question gave some statements about the Von Neumann model to correct, asked for another register in the fetch-execute cycle, and asked how an interrupt from an input device is handled.
The core idea. The Von Neumann model uses a single memory for both instructions and data, accessed over the system buses. The marks live in the precise role of each register — so know them exactly:
| Register | Full name | What it holds |
|---|---|---|
| PC | Program Counter | The address of the next instruction (not the instruction itself) |
| MAR | Memory Address Register | The address currently being read from or written to |
| MDR | Memory Data Register | The data being transferred to/from that address |
| CIR | Current Instruction Register | The instruction currently being decoded/executed |
| ACC | Accumulator | Results of calculations |
| Status register | — | Flags set by the ALU (carry, overflow, etc.) |
The two misconceptions that cost marks. First, thinking the PC holds the instruction — it holds the address of it. Second, mixing up the buses: the address bus carries the location, the data bus carries the data/instruction, and the control bus carries timing and command signals (it's the control unit that sends signals along it).
Handling an interrupt — describe it as an ordered process: at the end of each fetch-execute cycle the CPU checks for interrupts; if one is present its priority is compared with the current task; if it's higher priority the CPU saves its current state (register contents) to the stack, runs the relevant Interrupt Service Routine (ISR), then restores the saved state and resumes the cycle. That save-run-restore sequence is the mark-earner.
✏️ Practice it: "A printer signals that it has run out of paper while the CPU is mid-task. Describe, in order, how the CPU handles this interrupt and how it knows where to resume afterwards."
Number systems — binary, hex, two's complement & BCD
Another question tested converting a denary value to 12-bit binary and hex, interpreting a two's complement value, stating the 8-bit two's complement range, and giving a use for BCD.
Denary → binary → hex (the reliable method). Take any denary value and subtract the largest power of two that fits, repeatedly. Worked example with 213:
- 213 − 128 = 85, so the 128 bit is 1
- 85 − 64 = 21 → 64 bit is 1
- 21 − 16 = 5 → 16 bit is 1
- 5 − 4 = 1 → 4 bit is 1
- 1 − 1 = 0 → 1 bit is 1
That gives 11010101. To get hex, split into nibbles: 1101 = D, 0101 = 5 → D5. The exam trap is not padding to the required number of bits — if it asks for 12 bits, write all twelve, with leading zeros.
Two's complement → denary. If the most significant bit is 1, the number is negative. Method: invert all the bits and add 1 to find its magnitude. (For example, 11100010 → invert to 00011101, +1 = 00011110 = 30, so the value is −30.)
The 8-bit two's complement range is −128 to +127 — worth memorising, because the smallest is 10000000 and the largest 01111111. The asymmetry (one more negative than positive) catches people out.
BCD (Binary-Coded Decimal) stores each decimal digit in its own 4 bits. A good justified use is financial systems or digital displays: it represents decimal values exactly, avoiding the rounding errors you get when storing currency in ordinary binary floating point.
✏️ Practice it: "Convert denary 458 to 12-bit binary and to hexadecimal, then explain one reason a payroll system might store amounts in BCD rather than standard binary."
AI application & software licensing
The final question described a program that uses a camera to read printed words, translate them, and output audio — then asked how AI is used, two social benefits, features of a commercial licence, and why open source might be unsuitable.
Break the AI into its pipeline stages — that's exactly what separates a top answer from "AI makes it better":
- OCR (Optical Character Recognition) converts the camera's image of the text into actual characters.
- NLP / machine translation processes and translates those words, and can predict/complete text that's partially captured.
- Text-to-Speech (TTS) generates the spoken audio output.
Tie every point to one of those stages and the marks follow.
Social benefits should be real-world: it helps visually impaired users identify products or signs, and helps non-native speakers understand labels in a foreign language.
Commercial vs open source licensing:
- A commercial licence is typically paid, closed-source, protects the developer's intellectual property, prohibits legal copying/sharing, may limit the number of installs, and comes with official support and updates.
- Open source may be unsuitable here because the source code can be modified by anyone — and for a translation app that could introduce incorrect or unsafe output. A mistranslation of, say, an allergy warning could be genuinely dangerous, so the controlled, supported nature of a commercial licence matters.
✏️ Practice it: "Give one advantage and one disadvantage of releasing a medical translation app under an open source licence, and justify which licence model you'd recommend."
Quick recap
- Von Neumann marks are in precise register roles (PC = address, not instruction) and the bus distinction; interrupts follow check → save state → run ISR → restore.
- Master the denary/binary/hex/two's complement methods and remember the 8-bit range is −128 to +127; BCD avoids decimal rounding errors.
- Describe AI applications by their OCR → NLP → TTS stages, and know commercial vs open source licence trade-offs.
