NEC Computer Engineering common mistakes
NEC Computer Engineering · Understand the traps and how to avoid them.
Solving resistor networks by adding series and parallel values in the order they appear on the page instead of first redrawing the circuit into unambiguous series/parallel blocks.
Candidates pattern-match to the nearest formula and skip the reduction step; NEC-style questions embed bridge or delta-wye structures that look like simple parallel branches, so the arithmetic is right but applied to the wrong topology.
What to do instead
Before any calculation, redraw the network node-by-node, label every node, and mark which pairs share both terminals. If no pair shares both terminals, stop and apply delta-wye. Verify by checking that total resistance is bounded by the smallest branch and the simple sum.
Treating a Thevenin or Norton equivalent as valid for all load conditions and power calculations, including when the load is removed or when maximum power transfer is asked.
The equivalent circuit preserves terminal behaviour only for the linear external network; candidates forget the internal dissipation of the real network and compute load power from the equivalent source as if it were lossless.
What to do instead
Use the equivalent only for terminal voltage/current. For power in the actual components, return to the original circuit. For maximum power transfer, set load equal to Thevenin resistance and compute P = Vth^2/(4Rth), stating that this is load power, not source power.
Assuming an ideal op-amp in a circuit that is deliberately driven into saturation or operated open-loop, then reporting a linear output that exceeds the supply rails.
Virtual-short analysis is over-learned as a reflex; candidates never check whether the required output voltage lies within the supply range, so comparator and clipper questions are answered with impossible values.
What to do instead
After computing the linear output, always compare it with the positive and negative rails. If it exceeds a rail, clamp the output to that rail and re-solve the input node without the virtual short. State the operating region explicitly in the answer.
Applying a truth table or K-map minimisation to a sequential or asynchronous circuit as though it were purely combinational, ignoring state and feedback.
Digital questions often show gates with feedback or a state diagram; candidates reduce the gates statically and never consider that the output depends on stored state or that a race condition exists.
What to do instead
First classify: any feedback path or memory element makes it sequential. Build the next-state and output tables from the state encoding, then check for race, hazard, or lockout states. For asynchronous loops, verify the stable state for each input combination.
Computing cache performance with the wrong access-time model, e.g. adding hit time and miss penalty instead of using hit time plus miss rate times miss penalty, or forgetting that the tag check occurs on every access.
AMAT formulas are memorised as strings; candidates also confuse miss rate with miss ratio per instruction and forget write-back dirty-block traffic, so the answer is off by a factor tied to the miss rate.
What to do instead
Write AMAT = hit time + miss rate x miss penalty and define each term numerically before substituting. For write-back caches, add the dirty eviction fraction to the memory traffic. Sanity-check that AMAT lies between hit time and full miss time.
Answering pipelining questions with the ideal cycles-per-instruction (1) and ignoring structural, data, and control hazards plus the fill/drain of the pipeline.
The ideal CPI is the most memorable number; candidates forget that a k-stage pipeline needs k-1 extra cycles for a finite instruction stream and that stalls or flush penalties add cycles per branch.
What to do instead
Use total cycles = (k - 1) + n + stalls, and CPI = total cycles / n. Enumerate each hazard class, state the forwarding assumption, and add the branch misprediction penalty multiplied by misprediction rate and branch frequency.
Computing two's-complement range, overflow, or sign extension using unsigned rules, e.g. claiming an n-bit signed range of 0 to 2^n - 1 or detecting overflow by a carry out of the MSB.
Unsigned arithmetic is taught first and the carry-out habit persists; candidates do not distinguish carry-out from signed overflow, which occurs when operands of the same sign produce an opposite-sign result.
What to do instead
Memorise the signed range as -2^(n-1) to 2^(n-1) - 1. Detect signed overflow by comparing operand and result signs, not by carry-out. For extension, replicate the sign bit; for saturation, clamp at the range limits and state which rule applies.
Choosing a data structure or algorithm by name recognition rather than by the required operation mix, then misstating the complexity of the chosen structure's actual operations.
Candidates recall 'hash table is O(1)' or 'binary search is O(log n)' without the preconditions (uniform hashing, sorted array), and they conflate average with worst case, so questions about adversarial input or ordered traversal are missed.
What to do instead
Tabulate the required operations (search, insert, delete, ordered iteration, min/max) and the constraints (sorted? duplicates? memory?). Then give both average and worst case for each candidate. Never claim O(1) lookup without stating the hashing assumption.
Describing a communication protocol or bus at the level of 'it sends data' without naming the arbitration method, framing, error detection, or timing relationship that the question is actually testing.
Candidates study protocols as vocabulary lists and cannot map a scenario (collision, clock stretching, lost arbitration, CRC failure) to the mechanism that handles it, so they pick the plausible-sounding option.
What to do instead
For each protocol, build a four-row card: physical layer, arbitration/flow control, frame/error detection, and timing (synchronous vs asynchronous). Practise mapping a failure symptom to the responsible mechanism before reading the options.
Ignoring units, prefixes, and reference levels in mixed-signal and power questions, e.g. mixing dBm with dBW, RMS with peak, or 0 dB reference with absolute power.
The arithmetic is simple, so candidates rush and drop the conversion; dB questions in particular hide a factor of 10 or 3 dB, and RMS/peak confusion changes power by a factor of two.
What to do instead
Carry units through every step and convert to base SI before computing. State the reference explicitly (0 dBm = 1 mW). For sinusoids, write the peak-to-RMS relation before substitution, and check that a power answer is positive and physically plausible.