Nexus Learn

Lecture 3: Cost Analysis and Trends in Cost

Lecture Notes CS-704 Advanced Computer Architecture 7 min read Updated 8 Nov 2026

↓ Download notes (Word)

On this page
  1. Learning Outcomes
  2. §1 Trends in Cost
  3. §2 The Cost of an Integrated Circuit
  4. §3 Dependability
  5. §4 Common Mistakes
  6. Word Bank
  7. Self-Check Questions
  8. Further Reading

Learning Outcomes

SLO

After this lecture you can…

CLO

Bloom

SLO 3.1

Explain how the learning curve, volume and commoditization lower the cost of computers.

CLO1

C2 Understand

SLO 3.2

Calculate dies per wafer, die yield and the cost of a die for a given chip size.

CLO2

C3 Apply

SLO 3.3

Calculate MTTF, failure rate and availability of a system and show how redundancy helps.

CLO2

C3 Apply

§1.1 The learning curve

Slide L03-S06

The learning curve means manufacturing cost falls as a factory gains experience. The main measure is yield: the percentage of manufactured devices that pass testing. A new production line has low yield; as engineers find and remove the causes of defects, yield rises and each good chip costs less. Example: a new tandoor wastes many burnt rotis in its first week but almost none after a month - the same oven, more experience.

§1.2 Volume and commoditization

Slide L03-S07

  • Volume: a larger volume climbs the learning curve faster, makes buying and manufacturing more efficient, and spreads the one-time design cost over more units. Rule of thumb (Hennessy and Patterson): cost falls about 10% for each doubling of volume.
  • Commodity: a product sold in large volume by many vendors in nearly identical form (DRAM, flash, disks, keyboards). Competition pushes the price close to the cost. Example: USB drives and mobile chargers in any electronics market.

§2 The Cost of an Integrated Circuit

§2.1 From wafer to chip

Slide L03-S09

A wafer is a thin, round disc of silicon (30 cm across in modern factories). Many identical dies (the rectangular pieces that become chips) are made on it together. The wafer is tested, cut into dies, the good dies are packaged (placed in a plastic or ceramic case with pins) and tested again.

Roti analogy: Cutting square pieces from a round roti wastes the edges, and a burnt spot ruins whichever piece it falls on. A bigger piece is more likely to contain a burnt spot. That is why large dies cost much more.

§2.2 The formulas

Slide L03-S10

Cost of IC = (cost of die + cost of testing die + cost of packaging and final test) / final test yield

Cost of die = Cost of wafer / (Dies per wafer x Die yield)

Dies per wafer = pi x (d/2)^2 / A - pi x d / sqrt(2 x A)
(whole wafer) (partial dies lost at the edge)

Die yield = Wafer yield x 1 / (1 + D x A)^N

d = wafer diameter (cm) A = die area (cm^2) D = defects per cm^2
N = process-complexity factor (about 11.5 to 15.5 for 40 nm processes, 2010)

Source: Hennessy and Patterson (2012), §1.6. Assume wafer yield = 100% unless the question gives another value.

§2.3 Worked example: big die vs small die

Slide L03-S11

30 cm wafer, D = 0.031 defects per cm², N = 13.5, wafer cost $5,000.

Big die, 1.5 cm x 1.5 cm = 2.25 cm^2:
Dies per wafer = 3.1416 x 225 / 2.25 - 3.1416 x 30 / sqrt(4.5) = 314.2 - 44.4 = 269.7 -> 270
Die yield = 1 / (1 + 0.031 x 2.25)^13.5 = 0.40
Good dies = 270 x 0.40 = 108 Cost of die = 5000 / 108 = $46

Small die, 1.0 cm^2:
Dies per wafer = 706.9 - 94.25 / 1.414 = 706.9 - 66.6 = 640
Die yield = 1 / (1.031)^13.5 = 0.66
Good dies = 640 x 0.66 = 424 Cost of die = 5000 / 424 = $12

A die 2.25 times larger costs about 4 times more, because fewer dies fit on the wafer and each is more likely to contain a defect.

§2.4 Cost vs price

Slide L03-S12

Cost is what it takes to make one unit. Price is what the customer pays: cost plus research and development, marketing, sales, warranty and profit (together called the gross margin). NRE (non-recurring engineering) cost is a one-time cost such as design work and the set of masks used to print the chip; it is shared over all units sold, so a chip made for a small market is expensive per unit.

§3 Dependability

§3.1 Measures

Slide L03-S14

Term

Meaning

Formula / unit

MTTF

Mean time to failure: average working time before a failure

hours

MTTR

Mean time to repair: average time to fix after a failure

hours

MTBF

Mean time between failures

MTTF + MTTR

Availability

Fraction of time the system is working

MTTF / (MTTF + MTTR)

Failure rate

Failures per hour

1 / MTTF

FIT

Failures in time: failures per 10^9 hours

10^9 / MTTF

Example: a UPS works 2,000 hours between failures and takes 48 hours to repair. Availability = 2,000 / 2,048 = 0.977 = 97.7%.

§3.2 Worked example: MTTF of a system

Slide L03-S15

If failures are independent and each part has a constant failure rate, the failure rates of the parts add.

10 disks (MTTF 1,000,000 h), 1 controller (500,000 h), 1 power supply (200,000 h),
1 fan (200,000 h), 1 cable (1,000,000 h)

Failure rate = 10/1,000,000 + 1/500,000 + 1/200,000 + 1/200,000 + 1/1,000,000
= (10 + 2 + 5 + 5 + 1) / 1,000,000 = 23 / 1,000,000 per hour = 23,000 FIT
System MTTF = 1,000,000 / 23 = 43,478 hours (about 5 years)

§3.3 Redundancy

Slide L03-S16

Redundancy means having a spare part so that one failure does not stop the system. With two power supplies, the system fails only if the second fails while the first is being repaired.

MTTF(pair) = MTTF^2 / (2 x MTTR)
= 200,000^2 / (2 x 24) = 833,333,333 hours (about 4,200 times one supply;
Hennessy and Patterson round this to about 830 million hours)

Example: a home with both a UPS and a generator loses power during load shedding only if both fail together.

§4 Common Mistakes

Slide L03-S18

Wrong idea

Correct idea

A die twice as big costs twice as much.

Cost grows faster than area: fewer dies per wafer and lower yield.

Cost and price are the same.

Price = cost + R&D, marketing and profit (gross margin).

System MTTF = average of the parts' MTTFs.

Add the failure rates (1/MTTF) and take 1 / total.

MTBF and MTTF are the same.

MTBF = MTTF + MTTR.

Word Bank

Word

Simple meaning

Real-world example

Wafer

Thin round disc of silicon on which chips are made

A round roti before it is cut

Die

One rectangular piece of a wafer that becomes a chip

One square piece of the roti

Yield

Percentage of manufactured items that work

Good rotis out of 100 made

Defect density

Number of faults per cm² of wafer

Burnt spots per roti

Learning curve

Cost falls as a factory gains experience

A new baker improves in a month

Commodity

Product sold by many vendors, nearly identical

Mobile chargers, USB drives

NRE

One-time design and setup cost

Cost of making a new mould before printing toys

Gross margin

Price minus cost: pays for R&D, marketing, profit

Shop's profit on a phone

MTTF / MTTR

Average time before failure / time to repair

UPS working time / repair-shop time

Redundancy

A spare part to take over on failure

UPS plus generator

Self-Check Questions

  • What is the learning curve? How is it related to yield? (§1.1)
  • Why does volume lower cost? Give the rule of thumb. (§1.2)
  • What makes a product a commodity? Give two examples. (§1.2)
  • Write the formulas for cost of die, dies per wafer and die yield. (§2.2)
  • Find the dies per wafer for a 20 cm wafer and a 1.0 cm² die. (§2.3)
  • Explain why a die 2.25 times bigger costs about 4 times more. (§2.3)
  • What is NRE cost and why does it matter for small markets? (§2.4)
  • A server has MTTF 50,000 h and MTTR 10 h. Find its availability. (§3.1)
  • Find system MTTF for parts with MTTF 100,000 h (two of them), 20,000 h and 40,000 h. (§3.2)
  • Why does a second power supply increase MTTF so much? (§3.3)

Answers: 269 dies; availability 99.98%; system MTTF = 1 / (2/100,000 + 1/20,000 + 1/40,000) = 10,526 hours.

Further Reading

  • Hennessy and Patterson (2012): Chapter 1, sections 1.6 (trends in cost) and 1.7 (dependability).

← All CS-704 material