April marks the 20th anniversary of Tekmos' founding. Twenty years ago, James Betts and Lynn Reed left their jobs in California, moved to Austin, and started Tekmos. One of the more difficult things to do in a new company is to choose a name. We liked Tekmos, and when we googled it, there were no other hits, and we claimed it for ourselves.
Starting as a design center, we made the transition into a fabless semiconductor supplier two years later. We started by making digital ASICs, branched into Mixed Signal ASICs, and from there into standard products such as microprocessors, specialty memories, and peripherals. We also expanded from our original telecom emphasis into industrial, military, medical, and high temperature oil markets.
We just started work on our 830th chip. Looking back, it seems that our more interesting projects have all been related to either man-made or natural disasters.
In 2005, hurricane Katrina devastated New Orleans. This included the loss of the neighborhood phone switches. These switches relied on a custom chip that had been made at VLSI Technology, which had been acquired by Phillips, and whose fab had since been shut down. There was intense pressure to make 60,000 new chips, which were necessary for the full restoration of phone service in New Orleans. We got the job, since it had a mixed signal component, and so other companies declined to bid. The chip was a complicated one, with a lot of dual port RAM, and an embedded phase locked loop. We undertook a rush program to re-implement the design, and get it fabricated. Everything was expedited. This included working over the Christmas holidays, and putting in some very long days. In one event that stands out, we completed testing the first prototypes at about 2 AM on a Sunday morning. The customer purchasing agent was with us, and he arranged for a UPS pickup at that time, and the part to be flown directly to Mexico for assembly later on that day. We were surprised that such a thing could even be done. We delivered, and phone service was restored.
In 2011, the Tohoku earthquake occurred, which devastated the Japanese city of Sendai. Freescale had a fab in Sendai that was destroyed. The loss of that fab meant that they were unable to deliver what was to be a last time buy on microcontrollers to many customers. We set up an agreement with Freescale where we provided replacements for about 20 different microcontrollers. To do so, we designed a special gate array that contained ROM, EEPROM, and an ADC. This reduced our development time, and allowed us to switch devices part way through the manufacturing cycle if necessary in order to meet a particular demand.
One of the more difficult challenges was to recreate the 68020 microprocessor. The yield of this part was very sensitive to variations in poly sheet resistance, and the Sendai fab was the only Freescale fab capable of holding the required poly tolerances. And to complicate things, the original design documents had been lost. The only documentation available was the GDS data base and the production test vectors. All 51 of them. The chip had been designed using a combination of NMOS dynamic logic and CMOS static logic. Our design approach was to use our layout verification tool to extract the spice netlist from the layout. We then post processed the netlist to extract individual gates, and then to combine gates into higher order functions such as flops. We converted all dynamic logic into static logic, and re-implemented complex gates into collections of simple gates. And some of the complex logic gates were quite complex. Many of them had over 100 inputs. Then we began an iterative process, converting the production test into Verilog test bench vectors, and using them to check the design. Errors were corrected, and the simulations rerun to find the next error. When all of the errors were fixed. We made silicon.
We had many problems doing the 68020. The most difficult one was a case where the original design had transferred data onto a data bus, and left it there for a clock cycle. In essence, it was using the data bus as a dynamic 32-bit register. The circuit was not working, and we had a very hard time finding something that wasn't there. Once we realized what was going on, we had to add the register to make it work.
Not all disasters are caused by nature. Some are caused by men. One of our customers approached us with an obsolete FPGA that needed replacing. However, the original design files had been misplaced, and because of layoffs, the original engineers were gone as well. All that the customer had was the bitstream. In order to replace the design, we had to reverse engineer the bitstream. We would not do this if it were a case of someone trying to copy another design. But we will do it for the original owner. This was not an easy task. After deciphering the bitstream, we created a Verilog model of the FPGA, and loaded it with the bitstream. Then we used Synopsys to reduce the FPGA model to gates, which we then placed and routed.
This has been just a few of our stories. We have similar ones for each of the areas we operate in. Military stories, medical device stories, implantable stories, oil stories. Twenty years of stories. It has been fun and we look forward to the next twenty years.