Is Silverbrook's high speed Memjet printer a hoax?

Could Silverbrook's stunning Memjet inkjet be a hoax? This article looks at a few of the arguments from people who say that Silverbrook Research's impressive demonstration videos* of the one page per second inkjet cannot possibly be real.

Since this is an opinion piece, let's just put our cards on the table right away. Could Memjet be a hoax? No, we really don't think so. A great demonstration isn't the same as a finished product, we admit, but we believe the videos show fully working prototypes, just as Silverbrook Research has claimed.

We're convinced for three key reasons: the strong endorsement Silverbrook has received from Lyra Research, the leading imaging industry analysis firm; the track record of the scientists working at Silverbrook Research, particularly founder Kia Silverbrook; and lastly the towering stack of 3500 patents and patent applications that the company has piled up. We'll refer mainly to those documents in this article.

Yes, but...

Many people have pointed out that fast printing speed alone isn't enough to make a fast printer. The 1600 DPI printheads can produce two and a half million dots per square inch. You need to feed a huge amount of image data to the printer very quickly - about 220 Megabits per second for an A4 or letter-size page. If you can't, the printer will be slower because it will have to wait for the data.

Many more have asked how a fixed inkjet printhead with extremely narrow ink nozzles can cope with the clogging and blockages that seem to be a standard feature of inkjet printers.

We'll look at some of the technology that Silverbrook Research claims can overcome those problems.

By the way, this article will be much easier to understand if you've read some of Texyt's previous stories, which explain how the Memjet printhead is put together, examine the printing quality, and include a video of the prototypes.

More Inventions

Silverbrook's solution to both the challenges mentioned above depends on yet more inventions. One is the SoPEC chip (Small office home office Print Engine Controller). This is a computer on a chip, with a LEON SPARC CPU core, 2.5 megabytes of DRAM, USB 2.0 controllers, and interfaces for the printer's feed motor and printheads.

The SoPEC chip in the prototypes is an ASIC (Application Specific Integrated Circuit) designed by Silverbrook Research and manufactured by Magnachip at one of its fabs in South Korea*. See the photo on the right. Incidentally, the printheads are fabbed by TSMC in Taiwan, according to Lyra. (Update May 28: According to Memjet, the SoPEC chip is now being manufactured by IBM. Some sources claim that, although TSMC is involved in manufacturing, another Taiwanese firm will fab the micro-electromechanical part of the printhead).

“The driver chip is an integrated printing system on a chip that efficiently manages and processes large quantities of data, a page at a time, and drives the printheads in real time, calculating up to 900 million drops per second,” according to Silverbrook.

The SoPEC chip has three very interesting features.

1. A hardware JPEG decompressor

2. Two or more chips can be linked together to work together on the same print job

3. Firmware to compensate for 'dead' ink nozzles and misaligned print heads.

But 1600 nozzles per inch will get clogged up?

Before taking a closer look at the SoPEC controller chip, let's examine Silverbrook's answers to the problem of clogged ink nozzles on the print head. This is by far the leading cause of failure of current inkjets. If the printer doesn't get used often enough, the ink dries and sets hard inside the hair-thin nozzles. There's often no way to remove it without damaging the very delicate print head.

The Memjet A4/Letter size printer has more than 70,000 ink nozzles on its print head. Keeping all those nozzles unblocked sounds like an impossible task. Not only can the ink block them if it dries, but fine particles of paper and dust can, too.

Silverbrook has patented several different technologies that its researchers believe can deal with these problems. We don't know which of them are being used in the prototype printers.

Painting the sky in red and yellow

The amount of ink moving through each nozzle is so small that they may actually begin to dry out during printing if they're not used. Silverbrook has a way of dealing with this: make sure that all the nozzles get used every time a page is printed.

So your nice photo of a blue sky is going to end up spattered with thousands of red and yellow dots? Yes, that's correct.

But at 1600 DPI, they're too small to see as long as they're evenly scattered over the whole page, the company claims. Sounds unbelievable? Bear in mind that about 100 drops of ink from the Memjet print head would fit into the dot at the end of this sentence.

Doesn't this waste a ridiculous amount of ink? There are more than 70,000 nozzles on the print head, and we're going to shoot ink out of all of them every time we print a page!

Actually, the total amount of ink used if you fire every nozzle on the print head is about 0.00007 ml. That's because the ink drop size is only about one picolitre (a billionth of a milliletre). So you would have to print at least 15,000 A4 pages before you had wasted one ml of ink The Memjet A4/letter reference design calls for an ink cartridge holding 50 ml.

An ink drop is very cheap

So firing out thousands of extra ink drops is inconsequential. That's why Silverbrook has also patented a system for clearing nozzles which are already blocked, by warming them and pushing ink into them at high frequency. We'll take another look at that technology later in this article.

Further inventions to protect the printhead include a cap that covers it as soon as it stops printing. And there's also a pump that produces compressed air to blow any unwanted material away from it while it's printing. Silverbrook's patents even describe a system that uses the same air pump to push moistened air under the cap, to prevent the printhead from drying out when it's not used for a while.

We understand that compressed air may also be used to provide pressure to push ink through the long path from the cartridges through to the heating chambers that will fire the ink out of the nozzle and onto the paper.

SoPEC: one chip, 40 million transistors

With the sophisticated SoPEC chip, Silverbrook's printer concept has the CPU muscle to process all the incoming data on the fly, the company says. The maximum amount of data that needs to be fed to the printer, more than 200 Mb per second, exceeds the real world performance of typical USB 2 links, so compression is used. JPEG is well suited to squashing color photos and similar images.

JPEG isn't so efficient for text or line drawings. It appears that Silverbrook's driver figures out how best to compress different areas of a page, and uses a more suitable compression algorithm for text and line drawings.

The SoPEC ASIC has a lossless decompression module, or firmware, to handle this (unlike JPEG, such algorithms don't require complex floating-point calculations). The compression scheme used, SMG, is in fact a modified version of the group 4 compression used by fax machines. The printers do not print text by storing font files or transferring fonts from a host PC.

Since the two megabytes of memory on the chip won't always be enough to store the whole page, even compressed, the driver software feeds more compressed data to the printer as the page is emerging. The printer apparently cannot stop in the middle of a page and restart. We suspect this may be due to the risk of the ink nozzles drying out, among other issues.

The most interesting feature of the SoPEC chip is it's ability to compensate for dead ink nozzles. For example, the chip's firmware can boost some of the ink dots around a dead nozzle to partially disguise it. So, if a yellow nozzle is missing, it fires some extra yellow from adjacent yellow nozzles. 

Doesn't this mean you'll see a long, white line through the image? Probably not. Once again, an essential point to remember is that we're talking about a dot resolution of 1600 dots per inch, so that white line will be less than 16 microns wide – one seventh the width of a strand of hair, and it will be obscured by adjacent inkdrops spreading out. Silverbrook is aided by the fact that human color vision just isn't very sharp.

If you're familiar with the idea of hard disk drives or flash memory chips which mark bad sectors and stop using them, then this is similar. In fact, like some disk drives, the printhead appears to actually contain some spare nozzles, which are positioned to take up the slack when nozzles in front of them have failed.

A single SoPEC chip's dead nozzle correction unit has enough memory to handle cases where at least 5 percent of all the nozzles on the printhead are dead. So it seems that Silverbrook's researchers believe the printed output will still be acceptable with that level of nozzle failure or blockage.

How does the printer detect a dead nozzle? Well, strangely enough, it can't – according to some of Silverbrook's statements in patent applications - even though such a feature would seem to be an ideal way to deal with clogged ink nozzles and keep the printer running.

Several of Silverbrook's patents, rather oddly, go out of their way to assert that the print head doesn't contain any circuits to detect a faulty nozzle. This is very curious, because patents normally describe what an invention can do, they don't bother listing what it can't do. We've heard of companies being intentionally misleading in patents before, to confuse imitators. Could that be what's going on here?

What's particularly odd is that Silverbrook has patented a printhead design with individual thermal sensors on each nozzle, to detect overheating, and in fact, clogged nozzles do tend to overheat. Even more bizarrely, that very same patent also contains the company's apparently contradictory claim that the printhead contains no temperature sensor or blocked nozzle detector. More about this at the end of the article.

According to the patents, the dead nozzle compensation system seems to have two functions. One is to help handle printhead misalignment.

The Memjet printhead is made up of many 20 mm long sections, fitted together in a kind of jigsaw puzzle. The SoPEC chip can compensate for any that are slightly out of position. So-called 'dead' nozzles occur when there's a small gap between two sections of the head – a nozzle isn't where it should be.

We're going to speculate again, and guess that the printer prints out some test patterns after it comes off the production line, and these are used to fill in the dead nozzle map, and calibrate the controller chip to match the inaccuracies in the print head. It seems obvious that, if this happens, any genuinely dead nozzles will also be included. (Further examination of the patents since this story was originally written suggests that this guess is at least partly correct.)

This will greatly improve the production yields of the Memjet printhead fabrication process, since slightly faulty print head sections won't need to be discarded - and minor errors in assembly of the printhead can be compensated for.

We'll take another little step into speculation here and say 'why couldn't we use the same method to recalibrate the print heads after they've been in use for a while'? Perhaps when they begin to become blocked, and the printing quality deteriorates. A self-repairing print head? In fact, at least one of the patents does say that a printer's dead nozzle map can be updated at any time.

The print head that repairs itself?

We believe that if three of the Silverbrook Research inventions were combined, they might produce a self-healing printhead. They are

1. The thermal sensors on each nozzle
2. The blocked nozzle unclogging system
3. The dead nozzle correction unit

The thermal sensors can detect blocked ink nozzles. In the current Memjet inkjet design, a blocked nozzle will begin to heat up because it can't expel the heated droplet of ink it produces when fired.

The blocked nozzle unclogging system can respond to this by pulsing the nozzle at high frequency, to attempt to unblock it. The cleaning frequency used, incidentally, is 17KHz, which is very close to ultrasound cleaning frequencies.

If this fails, and temperature continues to rise, the SoPEC chip can add the nozzle to its list of dead nozzles, and stop using it. The SoPEC chip can compensate for the dead nozzle in future - it's designed to handle at least 5 percent nozzle failure on the print head.

However, although the technology appears to be in place, according to the patents, we were unable to find a patent or patent application where Silverbrook has linked these three systems together in this way. Although with 3500 to look through, we wouldn't be surprised if it's in there somewhere.

March 24: This article has been substantially updated with new information, corrections and clarifications.

March 31: Added note on dead nozzle map update capability mentioned in patents.

*Update April 2: Since this story was written, Memjet Technologies has removed all video of the wide format printer from its website. The company also removed another video which showed all four of its prototypes, including the wide format printer. However individual videos of the other three protoypes, an A4/Letter printer, label printer, and a photo printer, remain on the site. We don't yet know why Memjet removed the video of the wide format printer - see May 28 update also.

Update May 28: Memjet has added a new video of its wide format printer prototype to its site.

*We earlier incorrectly identified the chip maker, as either Hynix, from which Magnachip was spun off in 2004, or Hynix's former parent firm, Hyundai.