The emergence of high performance Thin Film Transistor Liquid Crystal Display (TFT LCD) technology over the last decade heralds a new era in desktop displays for computing applications. The long serving Cathode Ray Tube (CRT) display is in the process of being knocked off its perch as the premier display technology used in computing applications. Certainly the CRT will be with us for many years to come, since it is mature and cheap, but clearly the TFT LCD is the technology of the future.

The central issue at this time is price. The CRT despite its many limitations is highly price competitive due to the huge production volumes achieved, and the refinement in production processes which comes with half a century of mass production. In the simplest of terms, the CRT is to computer displays what the internal combustion engine is to automobiles.

Despite this, the clear performance and capability advantages of the TFT LCD have pretty much set the long term path for displays. The enormous production and R&D investment made by the Japanese and Koreans, in anticipation of an enormous consumer market in computer displays and High Definition Televisions (HDTV), clearly indicate that the CRT has only about a decade of life left in mainstream applications.

CRT Displays - The Technology Issues

The trusty CRT is the last piece of vacuum tube technology to survive in the mass production marketplace (excluding the specialised 2.45 GHz magnetron tube in your microwave oven). The central idea behind the CRT is to produce an electron beam from a heated cathode, accelerate that beam to tens of keV of energy, and drive the beam into a phosphor material which emits light when so excited. The beam, termed a cathode ray, is deflected magnetically (or electrostatically in some applications) in a raster scan pattern, to sequentially access individual pixels on the screen of the CRT. The beam is swept rapidly across a horizontal line - a row of pixels, typically at rates between 15 kHz (TV) and 100 kHz (hi-res monitors), then swung back to the beginning of the line to trace the screen again, while it is stepped down vertically, to address another row of pixels on the screen. Vertical stepping rates are much slower, between 50-60 Hz (TV) up to 100 Hz (hi-res monitors). By controlling the brightness of the beam as it addresses each pixel, we can paint a picture on the screen.

Now this mechanism alone provides only for a monochrome display, to achieve colour display three beams must be used, one for red, green and blue each, with each beam current controlled individually to achieve the desired hue, saturation and brightness desired for a given pixel. The beams each hit red, green and blue emissive phosphor dots or stripes on the face of the tube. A shadow mask, usually built as a thin sheet of perforated metal behind the screen, separates the three beams. Conventional tubes use triangular clusters of phosphor dots, whereas the popular Sony patented Trinitron uses stripes of phosphor.

Driving a CRT requires essentially five signals, three video channels which carry brightness for the individual beams, as an analogue voltage typically with a peak-peak swing of one Volt, and horizontal and vertical synchronisation (sync) signals which control the beam deflection circuits and hence, the addressing of pixels on the screen. Sync signals are typically TTL level.

A graphics adaptor for a CRT will typically store 8-32 bits of brightness information per pixel in a dual ported video RAM (VRAM) chip array. The CPU or accelerator hardware will write calculated brightness values into each pixel location via one port, while a RAMDAC (RAM controller and Digital to Analogue converter on one die) will extract the stored pixel value, decode it, and convert it to analogue voltages to directly drive the cable into the CRT display. Many RAMDACs also generate the sync signals, which may be separate or combined into a single TTL signal. Incompatibilities in sync signal formats, polarities and timing have long been used by manufacturers to lock competitors out of their markets, and the popular multisync monitor will only perform where the sync signal format is known and accommodated.

While purists may grumble at the superficiality of this treatment of CRT displays, it serves to illustrate both the simplicity of the CRT interface, and the basic interface model which is common to most LCD displays.

Other refinements to the interface have evolved in recent years, mainly to accommodate power saving and boot time probing, and a VESA standard exists to define additional serial TTL status monitoring and control signals between adaptors and CRT displays.

The strength of the CRT lies in maturity and cost, and potentially excellent picture contrast/brightness/resolution and viewing angle performance. In terms of picture quality per dollar, the CRT is hard to beat.

But the CRT also has it warts. Weight, volume and power consumption are all issues, and having a 21" monitor on your desk means abandoning all hope of using it for anything else useful. More serious failings also exist. One is the propensity of CRT guns to emit soft X-rays, ostensibly absorbed by additives in the tube faceplate glass. Nevertheless the potential for harmful X-ray exposure from prolonged use of CRT displays has led to a significant tightening of emission standards in recent years. The magnetic beam deflection system used also produces problems, since strong magnetic fields will exist around the neck of the tube, and picture quality is vulnerable to external magnetic fields. The profusion of CRT monitors in Australia with lopsided pictures testifies to a consumer base with no appreciation of the fact that a picture geometry adjusted in the Northern hemisphere's magnetic field won't do for the Southern hemisphere.

Other problems are also implicit in CRT designs. One is their propensity to produce nasty RF interference in the neighbourhood, by radiating amplified video and sync signals. The former effect, termed Van Eck radiation, can be sufficiently intense to allow a third party with suitable equipment to read the picture off your screen from tens to hundreds of metres away. Beware, if a nondescript panel van with antennas on its roof is parked in the neighbourhood !

Despite its failings the market for CRT displays has seen steady growth this decade, with worldwide USD 11B sales in 1993 growing to a projected USD 18B this year.Clearly the incumbency of the CRT will take some beating.

The LCD display is slowly penetrating the market, with 1993 sales of USD 2B growing to a projected USD 10-15B this year. We can expect the market to swing around the turn of the century, if current trends in LCD manufacture persist.

LCD Displays - A Technology Primer

The liquid crystal was discovered in 1888 by Renitzer, an Austrian botanist, but it was not until 1968 that RCA Labs in the US produced the first experimental display device. What makes liquid crystals useful for displays is their optical behaviour, since they posess a voltage dependent optical polarisation. Light rays passing through such materials are twisted through 90 degrees in polarisation, if no voltage is applied. If an electrical voltage is applied, the molecules in the liquid crystal change their alignment and the light rays are no longer twisted in polarisation.

If we sandwich a layer of liquid crystal between two glass plates, put a transparent sheet electrode on either plate, and then apply an electrical voltage of suitable magnitude to the electrodes, this sandwich of sheets becomes an electrically controlled optical polariser. If we then take two sheets of optical polariser, rotating one ninety degrees to the other, and place them either side of the liquid crystal/electrode sheet, we have a voltage controlled light valve. With no voltage applied, polarised light passes through the first polariser into the liquid crystal, is twisted by ninety degrees, and exits the second polariser. By applying a voltage to the liquid crystal, we force the molecules to untwist and the light rays retain their initial polarisation and are stopped by the second polariser. Voltage ON = Pixel OFF, Voltage OFF = pixel ON, and by varying the voltage we can smoothly control the amount of light passed though this sandwich panel. This is the basic physical model upon which all LCD display technology is based.

Liquid crystals can be driven with voltage levels of Volts to tens of Volts, and consume little power. However, they are slow to twist when voltage is applied, and this is a major problem for high performance displays intended to show rapidly changing images. This is the reason why LCD based calculator and photocopier displays were quick to emerge, but computer displays had to wait until the nineties.

To produce a matrix display suitable for bitmapped computer graphics, we need to divide the LCD panel electrodes into a matrix of individual rectangular cells, one per pixel (monochrome). By controlling the electrical voltage across each cell, we can control how much light is transmitted by that pixel. Typical displays use a flat fluorescent backlight behind the LCD sandwich, to provide a diffuse source of white light.

The most trivial arrangement for an LCD display is the Passive addressing scheme. Such arrangements use a row of vertical electrode stripes on one side to provide horizontal addressing of pixels, and a row of horizontal electrode stripes to provide vertical addressing of pixels. The pixel at the intersection of the row and column stripe electrodes with voltage applied to them is then addressed, i.e. turned on or off. We can then in a very simple fashion use a pair of counters to sequentially address each and every pixel, column by column in a row, and row by row, repeating the scan ad infinitum.

While appealing in its simplicity, this scheme in practice falls foul of the speed limitations of (Super) Twisted Nematic ((S)TN) liquid crystal materials, which require tens to hundreds of microseconds to switch state once a voltage is applied. Therefore we would cheat by addressing pixels by row, driving each column concurrently with a control voltage and using the horizontal electrode stripes to address row by row.

Let us assume that we are addressing a 640x380 pixel display with a horizontal frequency of 27 kHz and vertical frequency of 70 Hz. We have 37 microseconds to address each row of 640 pixels, assuming we drive all 640 columns with a specified analogue voltage, and we revisit each pixel every 14 milliseconds - this means that we have about 37 microseconds to drive each pixel, which has 14 milliseconds to wait before getting its next dose of electric field.

In practice passive schemes perform poorly, since the short time in which the voltage is applied produces poor contrast, narrow viewing angles and poor greyscale performance (i.e. a measure of how finely we can control the transmission through the pixel), little surprise since the pixel is being driven for less than 1% of the time. This is exacerbated by a propensity to crosstalk between pixels, as the electric field produced by applying voltage to a pixel can leak into its neighbour, and in effect partially drive it, smearing the image. For this reason passively addressed LCD panels never got beyond trivial resolutions.

As is evident, a more sophisticated scheme was required, to allow each pixel to be set to a voltage level and held there until the next addressing cycle. Ideally this could be accomplished by attaching a tiny capacitor to each pixel, and using a tiny switch to connect the capacitor to the column drive voltage once per addressing cycle. The TFT panel uses this method.

Active Matrix Thin Film Transistor LCD Panels

In a TFT panel design, Thin Film semiconductor technology is used to build a tiny transistor switch and capacitor for each and every pixel in the LCD panel. Using plasma deposition, sputtering, photolithographic and etching techniques similar in principle to those used in monolithic chip fabrication, transistors, capacitors and metal addressing lines are deposited on a glass substrate to form the required structures. Much of the machinery used for this task is derived from equipment used in conventional chip fabrication.

The most commonly used addressing architecture for TFT panels uses column "data" drivers to concurrently apply the required voltages to every pixel in a row, which is selected by a row "scan" driver. The scan driver turns the TFT switch on to charge the capacitor of every pixel in that row. Once the TFT switch is turned off, the capacitor holds the pixel at the set voltage (ie polarisation -> transmissivity -> brightness ) level until the next refresh cycle. Because the pixel matrix uses active TFT devices, it is termed an Active Matrix TFT LCD panel.

Because each pixel is isolated once the TFT switch is turned off, AM TFT panels have no pixel crosstalk problems, and isolation of the column lines from the capacitors and LCD pixel means that faster LC materials can be used, since there is not need for the inertia used in the passive architecture. Higher speed allows for much bigger panel sizes and resolutions, with superior contrast ratios, greyscale resolution and better control of viewing angles. The AM TFT panel is superior in every respect to the passive model.

While the use of TFTs solves many problems, it also creates problems, since it is by any standard a finicky fabrication process. The TFT switch is typically built as a MISFET (Metal Insulator Semiconductor Field Effect Transistor), mostly using Silicon Nitride as an insulator for the Gate, and doped Silicon for the transistor channel. The FET Source drives the Thin Film capacitor, the column voltage drive feeds into the FET Drain, while the Gate of the FET is connected to the selector scan drive. When the vertical scan drive voltage is applied to the gates of all TFT FETs in a row, they switch on and the voltage applied to the column drive is switched through the FET channel via the Source to the capacitor, charging it up. Readers familiar with DRAMs will appreciate the conceptual similarity.

TFT materials for AM TFT LCD panels are a science within themselves. The most common material used is amorphous Silicon (a-Si), common in Solar cell technology and decidedly inferior in electrical performance to the monolithic Silicon used in IC fabrication. The electrical conductivity and switching speed of a MISFET transistor both limit the addressing speed of a TFT panel, and critically depend upon the electron mobility (mu) of the transistor material. In this respect a-Si is woeful by any standards. An improvement can be made by using poly-Silicon (p-Si), which has two orders of magnitude higher mobility, allowing for much faster charging of pixel capacitors. However, conventional techniques for fabricating p-Si involve temperatures around 600 deg C, much higher than for a-Si, which tend to soften the cheaper glass substrates used with a-Si, forcing the use of expensive Quartz glass panel substrates. The latest technology for p-Si processing uses Excimer UV laser annealing of the transistor structures, to achieve the required p-Si properties without expensive glass substrates.

As is evident, the high cost of current AM TFT LCD panels is a direct result of very complex fabrication processes, which may produce often poor batch yields. The bigger the panel and its number of pixels, the greater the odds that a processing defect will occur rendering a pixel or row/column of pixels dead and thus resulting in an expensive and useless reject. In practice automated production machinery is used to process several panels together, with yields and throughput set by the size of the machinery. A line which produces a given defect density per square metre of panel, which processes a dozen 10" panels, will have a decidedly lower yield if it is used to make three 20" panels at a time. The reader is invited to calculate this, assuming a uniform density of defects per area !

At this point in the cycle we have made ourselves an AM TFT LCD panel by combining the aforementioned production processes and technologies. The panel alone however is not very useful without the necessary electronic driver circuits which push signals on to the panel's row and column addressing lines.

Driver Circuits for AM TFT LCD Panels

The electronics used to drive a TFT LCD panel are no less interesting than the innards of the panel. Two basic classes of device are used - Column Drivers and Row or Gate Drivers.

The simpler of the two is the Gate Driver, which is essentially a large register, with control and glue logic, which controls an array of high voltage analogue drivers. Essentially, the register propagates a "1" value along the rows in the TFT LCD panel, selecting one of the total number of rows, and driving it with TFT Gate control voltages with typical swings of 30-40 Volts. Since typical designs only cover 120-128 rows per chip, these devices can be cascaded to support vertical depths such as 768, 900 or 1024 lines by joining 6 or 8 Gate drivers respectively.

The Column Driver is by far the more complex circuit, since it has to latch a digital number determining the brightness of a pixel, and convert it to an analogue voltage to drive the TFT columns. A typical design will use a 9-36 bit single ended or differential TTL/CMOS bus for distributing the pixel values to the Column Drivers. The number of bits available per pixel determines the gray scale resolution of the display (and hence number of colours). Control signals from a synchronisation circuit are used to consecutively latch binary pixel values into the Column Driver. Once the pixel values are all latched, they are fed into a discrete Digital to Analogue converter (DAC) to produce drive voltages for each an every column. Due to LCD non-linearity a Gamma-correction scheme is often used, with the voltage levels for each Greyscale level produced by an external analogue resistive divider circuit. In this manner, specific control can be exercised over the display's Gamma correction. It is worth noting that the discrete drive levels used simplify the DAC design down to an analogue mux. Finally the output from the DAC is fed into drives which feed the TFT columns. Like Gate Drivers, the Column Drivers are typically cascaded, with 300-309 or 384 columns driven per driver. By appropriately combining multiples of these values, common resolutions such as 1024, 1152, 1280 or 1600 columns may be produced.

Some measure of the internal throughput on a column driver pixel bus can be gained from the typical clock speeds of 65 MHz and widths of 9 to 36 bits, producing bus throughputs of 73-293 Mbytes/sec. Not a trivial number, by any measure.

Contemporary TFT LCD drivers are typically made using conventional monolithic IC fabrication methods, and packaged using Tape Automated Bonding (TAB) techniques for low cost. The TAB packaged drivers are glued on to the glass substrate around the edges of the panel and the leads are connected via one of many possible techniques to the metal column and row leads fabricated on the glass. It is hoped that the next generation of p-Si based technology may allow the driver circuits to be made of TFTs and directly fabricated on the glass substrate with the rest of the panel. Significant cost savings will then follow.

The internal interface to the driver circuits for a TFT LCD panel, as used for instance in a laptop, can be wholly digital, bypassing the need for a RAMDAC in the graphics adaptor, as is used in 99% of contemporary desktop computers. Basically a bus is required to carry data for each pixel, and a clock and sync signals are needed to tell the panel drivers where to put each pixel value.

Should the TFT LCD panel be used in a monitor, then an Analogue Digital Converter (ADC) will be needed to digitise the incoming analogue RGB video so it can be placed upon the pixel data bus. Unless an expensive ADC is used, some loss of colour resolution may occur. To complete the monitor design, typically a microprocessor will be embedded to control the setup of the monitor, using Non Volatile RAM or EEPROM to save Gamma correction and brightness, contrast and colour balance.

The observant reader will note that up to this point we have not explained how colour is produced. This was intentional.

The basic technique for making a colour AM TFT LCD is to simply triple the number of TFT/capacitor/electrode cells (until now referred to as pixels), and produce a three colour RGB optical filter layer, so that each pixel is really made up of three individual subpixels, one for each colour, and additional Column Drivers to select these. A rectangular pixel is split into three subpixels, each with a coloured transparent rectangle per subpixel. Monstrously complex ? Indeed, since we have just had to either triple the number of Column Drivers or build in a clever multiplexing scheme so that we can select which colour cell we are to drive.

AM TFT LCD Monitor Products

The market for high resolution AM TFT LCDs has steadily grown in recent years. The first volume market for these displays was in military avionics, since "glass" cockpits are the current trend and price was not an issue where weight and volume are a high priority. These early displays had sizes of about 6" and resolutions between 512 and 800 pixels square. The technology quickly proliferated into laptops, as Japanese and Korean manufacturers mastered 8-10" 800x600 and 1024x768 resolution displays. We are now seeing rapid growth in the monitor market, with the wide availability of 10", 12", 13.8", 14", 14.5" and 15" panels offering resolutions of typically 800x600 or 1024x768 pixels with 16-bit colour resolution.

The latest generation of production monitors have sizes of 16.1" and resolutions of 1280x1024 pixels at 16-bits. We can expect at least one 21.3" 1600x1280 16-bit colour monitor to enter volume production this year, with several manufacturers claiming 20-21" class products in the pipeline.

Because LCD panels utilise the whole screen size for the picture, the diagonal size can be deceptive in comparison with CRT sizes. A 14" LCD provides similar viewing area to a 16-17" CRT monitor. Therefore a 21.3" inch LCD is closer in size to a 24" CRT.

I had the pleasure recently of using a 14" LCD monitor and can happily state that the assertions of the vendor community are mostly correct, since the device provided picture quality comparable to a top end CRT, and much better picture sharpness and geometry. At roughly three times the cost of a CRT, it is expensive viewing but for many applications where space, weight, power consumption and eye fatigue are an issue, arguably a justifiable expense.

The OEM AM TFT LCD monitor typically uses standard analogue VGA/SVGA/XGA/SXGA 15 pin I/O, dissipates 30-50 Watts and is a direct drop in replacement for a CRT monitor, with the caveat that speeds and resolutions will be restricted in comparison with a multisync CRT. That aside, the only impediment to wider market penetration is and will continue to be cost.

The TFT LCD Monitor is here to stay, and without any doubt the best is yet to come.