IELTS Academic Reading Practice Test 206

In 1796 a young Bavarian writer named Alois Senefelder, short of money and looking for a cheap way to print his own plays, discovered a principle that would still be shaping technology two centuries later. Senefelder found that if he drew on a smooth slab of limestone with a greasy crayon and then wetted the stone, printing ink would cling only to the greasy lines. The stone did not need to be carved: the image and the blank areas lay in the same flat surface, separated by nothing more than a chemical difference. He called the process lithography, from the Greek words for stone and writing.

For a hundred and fifty years lithography remained chiefly a way of printing pictures and posters. Its underlying idea, a pattern defined by chemistry rather than by physical relief, seemed to suit a quite different problem that appeared in the 1950s. Electronics was moving from bulky valves to transistors, and manufacturers needed to make thousands of identical devices on a single slice of semiconductor and to connect them without wires.

The first integrated circuit was built in 1958 by Jack Kilby at Texas Instruments, using germanium. A few months later Robert Noyce, working at Fairchild Semiconductor, proposed a more practical version built on silicon. Its key feature came from a colleague, Jean Hoerni, whose planar process covered the silicon with a thin protective layer of oxide. Openings could then be cut through that layer wherever connections or new materials were needed. The oxide was a mask, and the problem was how to cut the openings precisely and cheaply.

The answer was photolithography. A wafer is first coated with a thin film of a light-sensitive polymer known as photoresist. Above it is placed a mask: a transparent quartz plate on which the pattern for one layer of the circuit is formed in a thin layer of chromium. Ultraviolet light is shone through the mask, and wherever it reaches the resist the polymer changes chemically. The wafer is then washed in a chemical developer, which dissolves the resist that has been exposed. The oxide beneath now lies uncovered in exactly the required pattern, and is removed by etching. Finally the remaining resist is stripped away, leaving a patterned surface ready for the next layer.

Each chip needs dozens of such layers, stacked with extraordinary precision, so the machinery that transfers the pattern has been refined relentlessly. The earliest aligners pressed the mask directly against the wafer. This contact gave sharp images, but every contact risked scratching the mask and adding a speck of dust that would ruin the chip, and the mask wore out quickly.

Projection solved that problem by keeping mask and wafer apart. In 1973 Perkin-Elmer introduced the Micralign, which used mirrors rather than lenses and swept a narrow slit of light across the wafer. Later in the 1970s the American firm GCA developed the step-and-repeat aligner, or stepper. It exposed a small area, moved a fraction of a millimetre, and repeated the pattern across the whole wafer. Steppers also reduced the image, typically to a fifth of its size on the mask, which eased the burden on mask makers.

Resolution depends above all on the wavelength of the light used. For years machines relied on mercury lamps, whose strongest lines lie at 436 and 365 nanometres. As features shrank, manufacturers turned to excimer lasers, which produce intense ultraviolet light from excited molecules of gas. Krypton fluoride gave 248 nanometres and argon fluoride 193, and the industry rebuilt its optics around each new wavelength in turn. Shorter waves meant finer lines, but also new glasses, new coatings and new resists.

By the late 1990s a wall appeared. The next obvious step, a laser producing 157 nanometres, was abandoned because no lens material of sufficient quality could be produced in the quantities required. Engineers took a different route instead. If the space between the final lens and the wafer is filled with ultrapure water, the light behaves as though its wavelength were shorter, because water bends light more strongly than air does. Immersion lithography, as this became known, was introduced in the early 2000s and prolonged the working life of 193-nanometre equipment by two decades.

The latest answer is extreme ultraviolet, with a wavelength of 13.5 nanometres. Light at that wavelength is absorbed by air and by every known lens material. The process must therefore take place in a vacuum, and the beam must be steered by mirrors with atomically smooth coatings. The light itself is made by firing a laser at tiny droplets of tin, which turn into a plasma hot enough to radiate at the required wavelength. Only one company, ASML of the Netherlands, has succeeded in building such machines for sale, and each costs well over a hundred million dollars.

In 1965 Gordon Moore, later a co-founder of Intel, observed that the number of components on a chip had been doubling at a steady rate and predicted that it would continue. Revised in 1975, the observation became known as Moore's law. It is not a law of physics but a target that the industry has organised itself to meet.

Anyone who has used email for long enough will have received a message that claims a fortune is waiting, offers a remedy that no pharmacy sells, or simply repeats a brand name at length. Such messages are known as spam, a term borrowed from a television comedy sketch of 1970 in which a café serves every dish with the same tinned meat and a chorus of customers drowns out all conversation. The word captures the essential quality of the phenomenon: it is not one unwanted message but a flood that makes the useful ones harder to find. In the mid-2000s, industry estimates suggested that the great majority of all email traffic was spam, and although the proportion has fallen, the volumes remain enormous.

The first defences were lists of rules. A filter would give a score to words and phrases that experience had linked to unwanted mail — 'free', 'winner', 'act now' — and would block a message once the total passed a threshold. Rules could also be written by the user, so that mail from a particular address was always accepted or always refused, though such lists of approved senders demanded constant attention as contacts changed. The filters were cheap and transparent, and they still run on many servers. But they were also easy to defeat. Sending 'fr3e' or 'w1nner' cost a spammer nothing and slipped past a list that had not foreseen the spelling. A filter could also stop a message the user genuinely wanted, simply because it discussed a competition the reader had entered.

A more robust approach was to block the sender rather than the words. In the late 1990s a system called the Realtime Blackhole List, devised by Paul Vixie, published the addresses of computers known to send spam. A receiving server could ask a special database whether the machine connecting to it appeared on the list, and refuse the message if it did. Shared blacklists of this kind are still widely used. Their weakness is that they punish the computer rather than the person. Once spammers began to send their mail through thousands of ordinary home computers that they had infected with malicious software, the machines changed from hour to hour, and blocking them one by one became futile.

Attention therefore turned to proving who had sent a message. Three related standards now do most of this work. The first, Sender Policy Framework, allows the owner of a domain to publish a list of the computers permitted to send mail on its behalf, so that a receiving server can check whether a message claiming to come from a bank really did. The second, DKIM, attaches a cryptographic signature that the receiver can verify. The third, DMARC, tells the receiver what to do when either check fails. None of these systems prevents a determined sender from registering a new domain; what they do is make forgery detectable and give reputation systems something reliable to work with.

The most important change, however, came from statistics. In 2002 the programmer and essayist Paul Graham published a proposal for filtering mail with a mathematical technique known as Bayesian classification. The idea was to train a filter on each user's own correspondence: messages the user had marked as spam, and ordinary messages they had kept. For every word and combination of characters, the filter calculates how much more often it appears in one pile than the other, and then combines those probabilities into a single score for each arriving message. Because the training is personal, the filter learns that 'invoice' is ordinary in an accountant's inbox and suspicious in a teenager's.

Bayesian filters worked remarkably well, and they were soon built into most email services. Their weakness is that they treat each clue as independent, which is mathematically false: 'free' and 'offer' tend to arrive together. In practice the error hardly matters, because the combined evidence is strong. Spammers responded by padding their messages with respectable text, taken from books or news reports, to dilute the suspicious words, or by hiding their message in an attached picture, where a text filter cannot read it. Filters then learned to read the text inside images, and to compare an image with thousands of known spam images.

Modern systems combine all of these signals and add others. Reputation services score the history of the sending machine and of the domain. A technique called greylisting makes use of the fact that legitimate mail servers, if told to try again later, will do so, whereas a machine that is sending millions of messages will usually not bother. Greylisting is popular because it is cheap, but it delays every first message from a new sender by several minutes, which some users find irritating. Filters also examine the structure of a message — the order of its headers, the software that generated it, the time of day it was sent — because bulk mail carries traces of the program that produced it.

No filter is perfect, and the two kinds of error do not cost the same. Marking a genuine message as junk, a false positive, may lose a job offer or a hospital appointment; letting spam through merely wastes a second. Systems are therefore tuned to err on the side of caution, which is why a message from an unknown sender sometimes never arrives. The arms race continues: as authentication becomes normal, spammers concentrate on compromising legitimate accounts, whose messages pass every check. Some researchers argue that the most useful advances now lie less in better mathematics than in persuading more organisations to adopt the standards that already exist. Yet the economics favour the attacker, since sending a million messages costs almost nothing and even a negligible response rate can be profitable.

A death now leaves two kinds of estate. The first is familiar: a house, a bank account, a box of photographs. The second is invisible to almost everyone except the companies that hold it: an inbox containing a decade of correspondence, a social media profile, a library of pictures stored in the cloud, and the questions a person typed into a search engine. This second estate may outlive the first by many years, and in most countries the law has almost nothing to say about it.

The reason is that an account is not property in the ordinary sense. What the user holds is a licence to use a service, granted under a contract that few people read and that usually says nothing about death. Providers have filled the gap individually. Facebook allows an account to be turned into a memorial page and, since 2015, has let users nominate a legacy contact who can manage it; Google's Inactive Account Manager, introduced in 2013, offers to notify chosen people, or delete the account, if it goes unused for a set period. Others release the contents of an account only on the order of a court.

Privacy law offers little help, because it is written for the living. The European Union's General Data Protection Regulation states explicitly that it does not apply to the personal data of deceased persons, and the United Kingdom's legislation is framed in the same way. A will can dispose of possessions, but it cannot easily dispose of a password, and in many jurisdictions the right to read a dead person's messages does not pass to their heirs. The result is that the fate of a considerable part of a person's life is settled by the terms of a contract they accepted with a click.

The consequences are not merely sentimental. Photographs may be the only record of a childhood, and their loss is permanent in a way that the loss of a paper album is not: there is no negative to find. Financial records may be needed by those administering an estate; messages may contain evidence in a dispute; medical data held by a fitness app may matter to a surviving child. And there is a darker side. Identity thieves have been known to use the accounts of the recently dead, a practice that succeeds because the companies holding those accounts hear nothing about the death.

Companies, meanwhile, have their own stake in the data. A profile that continues to attract advertising is worth money. It would be unfair to assume that providers behave cynically; many have built thoughtful tools, and it is rarely in their interest to refuse a straightforward request. But when a company's interests and a family's wishes diverge, there is no neutral party to decide.

A newer industry has made the question harder still. Several firms now offer to keep a version of a dead person speaking: a chatbot trained on their messages that can answer new questions in something resembling their style. The technology is crude, and the resemblance is often superficial, but the commercial appeal is powerful. Whether such a creation is a comfort or an indignity is not a question that can be settled by engineering, and the person whose voice is copied is in no position to object.

The case for stronger rules rests on a simple point: a person's wishes about their own traces deserve the same respect as their wishes about their money. If I can leave my house to my sister and my savings to my brother, why can I not leave my photographs to one and deny them to another? A legal status for a digital executor, someone empowered to carry out instructions recorded in advance, would give effect to choices that at present are often ignored. Providers should be required to state plainly what happens on death, and to offer a default that a user can change.

The case against regulation is not contemptible. Data about one person is rarely only about that person: an inbox is a record of other people's confidences, and a photograph may show someone who never agreed to its preservation. Handing over an entire archive to a relative may protect one person's wishes and violate another's privacy. There is also a risk that poorly drafted rules would freeze useful material, since historians and epidemiologists rely on the traces of ordinary lives. And a dead person cannot give informed consent to anything, so every scheme is a substitute for consent rather than the real thing.

None of this persuades me that the present arrangement is defensible. The asymmetry is too great: the company knows exactly what will happen, the user knows nothing, and the family discovers the rules only when it is too late to act. My own view is that deletion should be the default, applied after a reasonable interval unless the user has chosen otherwise, and that the choice should be made when the account is opened rather than buried in a settings menu. A named executor, wider use of tools that already exist, and a legal duty to explain the terms in plain language would cost companies little and spare families a great deal.

Death has always required the living to make decisions on behalf of the dead, and the law has developed careful procedures for doing so with property. What is new is the scale and intimacy of what is left behind. It is time the same care was applied to the traces we leave online.