Hardware Architecture of Texas Instruments' Electronic Calculators

Even before Jack Kilby's Invention of the Integrated Circuit (IC) in July 1958, enabling Electronic Calculators as we know them today, there were already Digital Computers. While the concept of Boolean algebra, the representation of the values of variables with "true" and "false", usually denoted with 1 and 0, was known since 1847, was it Konrad Zuse developing and building from 1936 to 1938 with the Z1 the World's first freely programmable computer using binary numbers. To overcome the unreliability of this motor-driven mechanical computer, the design of the first "Electronic Numerical Integrator and Computer" better known as ENIAC, started in 1945 and was based mainly on vacuum tubes and crystal diodes for its logic. Next step of evolution, often called second-generation computer, was the introduction of fully transistorized computers mid of the 1950s and laying the foundation for Electronic Desktop Calculators. In 1964 three fully-transistorized desktop calculators arrived on the market within a few weeks: Friden EC-130, IME 84rc, and Sharp CS-10A. The EC-130 used a Cathode Ray Tube (CRT) to display 4 lines of a calculation and both IME 84rc and CS-10A used Nixie tubes, a cold cathode display, to output up to 16-digit decimal numbers in one line. Sharp's CS-10A design used a total of 530 germanium transistors and 2,300 germanium diodes and sold in 1965 for around USD 2,500 or almost US$ 25,000 in 2023 money.

The first commercially available IC, the SN502 Solid Circuit Flip Flop introduced in March 1960 for US$ 450 per unit, was not an immediate option to reduce manufacturing costs of Electronic Desktop Calculators but paved the way toward compact, reliable and affordable product. Starting in 1961 different Digital Circuits manufactured in bipolar silicon technology were introduced to the marked based on RTL (Resistor-Transistor Logic), DTL (Diode-Transistor Logic) and TTL (Transistor-Transistor Logic) leading to higher and higher complexity of the ICs and resulting in a very fast product innovation cycle. Looking in Sharp Corporation's Calculator Innovations, you'll find:

When Jack Kilby started in 1965 together with his colleagues Jerry D. Merryman and James H. Van Tassel the Cal-Tech Project, they envisioned building an IC-based, battery-powered "miniature calculator" that could add, subtract, multiply and divide, yet could fit in the palm of the hand. Using large and power-hungry Nixie tubes wasn't an option for the team to display the calculating results and they invented a technology using a dot-matrix heat element printing serially on a roll of heat-sensitive paper. After two years of development, the three TIers completed the first hand-held calculator in 1967. The battery-powered device could accept six-digit numbers, perform the four basic arithmetic functions, and print results as large as 12 digits on a thermal printer. The calculator had a case fashioned from a solid piece of aluminum; it measure about 4-1/4 by 6-1/8 by 1-3/4 inches and weighed 45 ounces. The Cal-Tech project was meant as "Proof-of-Concept" for so-called Large Scale Integration of Integrated Circuits and it went well. Canon commercialized the technology developed by Texas Instruments and introduced in 1970 with the Pocketronic the World's first battery-powered printing calculator.

In 1965, Gordon E. Moore - co-founder of Intel - postulated that the number of transistors that can be packed into a given unit of space will double about every two years. This allowed to integrate more and more transistors, diodes and resistors into Integrated Circuits and we differentiate between "Small Scale Integration" (SSI) with complexities in the lower tens of components, "Medium Scale Integration" (MSI) with tens to hundreds of components on a single chip and "Large Scale Integration" (LSI) with one thousand and more components integrated on a small silicon chip. Looking again into Sharp Corporations' Calculator Innovations, we basically can see that around 2,800 discrete components were reduced with the next generation of desktop calculator to around 2,100 discrete components and 28 ICs (SSI) and the following generation used only 72 ICs (SSI and MSI). Next major milestone for Sharp was the introduction of the QT-8D in 1969, reducing the complexity of the design further to 6 ICs (4 LSI and 2 MSI):

The design of LSI ICs was mainly a result of switching from bipolar transistor technology to Metal-Oxide Semiconductor (MOS) technology, allowing higher densities for transistors used with digital functions like logic gates and storage registers found in electronic calculators and computers. In 1969 it was commercially possible to integrate about 1,000 transistors in a p-Channel MOS (PMOS) process on a silicon die measuring about 0.2" x 0.2" (5 mm x 5 mm) and having about 28 to 48 electrical connections to the outside world. Using the equivalent of around 5,000 transistor functions for a typical electronic calculator design, most "Chipsets" for electronic calculators introduced around 1970 consisted of about 3 to 6 LSI ICs.

Looking closely at the silicon die of the TMS1000, the first commercial available Microcomputer introduced in 1974 with the SR-10 calculator, you will notice that two larger blocks on the die use about 50% of its real estate: The 256-bit Random-Access Memory (RAM) at the upper left of the chip and the 8,192 Bits Read-Only program Memory (ROM) at the lower left of the chip. In 1965 every transistor counted and neither RAM nor ROM was an option for the earliest electronic calculators. How do electronic calculators work? Remember how to add two 2-digit numbers with pencil and paper? Yes, align the numbers so that the tens' and ones' line up neatly and draw a line under them. Add the digits in the ones' column together, if the result is larger than 10, add a one to the tens' column. Finally add the numbers in the tens' column together and if the result is larger than 100, add a one to the hundreds' result. And multiplications and divisions are just simple algorithm going back to a large numbers of additions and subtractions, respectively. This is precisely how the first electronic calculators worked. The numbers are represented as "Binary-Coded Decimals" (BCD), meaning 4 bits are grouped together for each digit of the number with the individual bits having a weight of 8 (2³), 4 (2²), 2 (2¹), and 1 (2⁰):

Using now a 4-bit Binary Adder with two BCD numbers results for example with 3 + 5 as operands to the expected result of: 0b0011 + 0b0101 = 0b1000 or decimal 8. But trying to add 5 + 7 results in 0b0101 + 0b0111 = 0b1100 which is not defined as a BCD number. BCD-Adders detect this kind of overflow and correct the result by adding the constant 6 to the interims result, giving 0b1100 + 0b0110 = 0b10010 with the leftmost bit as a carry over into the digit to the left and finally resulting in 0b00010010 or decimal 12. Adding now two large 12 digit numbers would not necessarily require a 48-bit Binary Adder, the numbers are added digit by digit, from the right to the left. Actual implementations sometimes even substitute the 4-bit Adder with a 1-bit Adder used 4-times in a loop per digit. Hewlett Packard's innovative HP-35 is representing numbers with 56 bits, divided into 4 bits for the sign of the mantissa, 40 bits for the mantissa, 4 bits for the sign of the exponent, and 8 bits for the exponent and using a 1-bit Adder for a bit-serial architecture versus digit-serial architectures using a 4-bit Adder.

The above mentioned 256-bit RAM of the TMS1000 would be capable of storing up to 4 numbers of 16-digit length but has an extremely high transistor count due to its design and implementation. Early transistorized calculators used a completely different approach to store numbers, Friden's EC-130 for examples used a magnetostrictive delay line, an acoustic delay line technology tracing back to the mercury delay lines pioneered in the 1950s, a truly serial type of storage. The electronic equivalent of delay lines are shift registers and with the advent of MOS LSI technology, shift registers were actually the first mass-produced PMOS chips starting in 1964 with 20 bits and leading within a few years to densities of hundreds of bits. Texas Instruments Cal-Tech demonstrator used for storage of the numbers three shift registers and divided the remaining electronics into four logic chip and their approach proved correct, dynamic and later static shift registers dominated the design of electronic calculators in their infancy, hitting the sweet spot of high integration density and application match of serial architectures. The resulting memory configuration is called Serial Access Memory (SAM) and the numbers are stored like on a racetrack and always circling, contrary to RAM where every bit can be accessed in a random order.

The other large block of the TMS1000 is its 8,192 Bits ROM to store the program of the calculator, like scanning the keyboard, controlling the display, managing the registers for the adders and many more tasks. Well, there are differences between computers and calculators and early electronic calculators didn't use any program memory. In the early days of Electronic Calculators, it was all about transistor count - with fully-transistorized calculators it defined the manufacturing costs, product size, power consumption, reliability and competitiveness while with LSI technology it defined how many (expensive) PMOS ICs the calculator design would need. To reduce transistor count to the bare minimum, their was no programmability at the beginning - everything was "hard-coded" in the chip design. Advantage of low transistor count was disadvantage of inflexible designs and following Moore's Law within a few years two major changes took place:

Here at the Datamath Calculator Museum we differentiate for calculators using Register Processors six steps of evolution from the Cal-Tech design:

The flexible design concept of the TMS0100 with both programmable PLA and ROM techniques allowed for easy individualizations and hundreds of pocket calculators and small desktop calculators used one of its 28 known chip variations and the groundbreaking modular approach of the TMS0200 Building Blocks was further optimized with the TMC0500 Building Blocks leading to the TI-59 and being remembered for eternity. The eclipse of TMS0200 and TMC0500 Building Blocks was clearly the SR-60 Programmable Prompting Desktop Calculator with integrated Printer introduced in 1976 and using in its base configuration fifteen Building Blocks. The last Register Processor developed by Texas Instruments was the TMC1500 introduced in 1977 and not only integrating the equivalent of three TMC0500 Building Blocks but the Display Drivers, too on a small silicon die and featuring about 30,000 transistors.

But the Register Processors inherited a disadvantage from its roots tracing back to the earliest, fully-transistorized Electronic Calculators - the shift register based Sequential Access Memory. While the sequential access of the individual digits of numbers represented in BCD format works perfectly to add large numbers, did it block the use of the chips for other applications. Single-chip calculators could add numbers but they could not keep time, control a furnace or do anything else - like a flexible microcomputer could do.

Texas Instruments realized these limitations rather quickly and the successor of the TMS0100 "calculator on a chip" was announced in 1974 as TMS1000 "computer on a chip". Comparing both the TMS0100 and TMS1000 reveals many similarities but one massive difference: The 182-bit Serial-Access Memory (SAM, 3 Registers * 13 Digits, 2 * 13 Bit-Flags) was replaced with a 256-bit Random-Access Memory and the 3,520 Bits Read-Only program Memory (ROM, 320 Words x 11 Bits) was increased to 8,192 Bits (1,024 Words x 8 Bits) and the TMS1000 Microcomputer Family took the World in a storm and was the most successful 4-bit design ever. Well, designing calculators with a Digit Processor introduced new challenges! Register Processors could add two large BCD numbers with just one instruction on the fly and Digit Processors literally need to pick one digit from one operand, one digit from the other operand, add them together, store the result back, load the next digits form the first operand and so on. Sounds time consuming? It is time consuming and Texas Instruments learned its lesson the hard way when trying with "Project X", started 1977 and abandoned in 1982, to replace the TI-59 based on a TMS0501E Register Processor introduced in 1974 with the TI-88 based on two TP0485 Digit Processors. But here Moore's Law is fortunate, too. Integrated Circuits did not only double their transistor count every two year, their clock frequency and operating speed is increasing at an almost similar rate and we are not aware of any Register Processor developed after the 1977 by Texas Instruments. Switching to flexible Digit Processors allowed designers suddenly to work not only with serial adders on long BCD numbers, they could do completely different operations on any data that could be represented and stored in the RAM of the devices like the famous Little Professor Educational Toy, the groundbreaking Speak & Spell or the TI-2000 Time Manager.

The ongoing battle of adding new features to the next product before Moore's Law allowed Single-Chip solutions, created with Digit Processors some Multi-Chip designs, most notably architectures using a Main-CPU and one or more Support-CPUs. A very good example is the TI-5050 Printing Calculator introduced in March 1975 and using two Digit Processors of the TMS1000 Series, one for keyboard scanning, calculations and display output and a second one for the thermal printer. The TI-5050 was replaces after only 15 months with the TI-5050M, adding a User Memory to the feature set of the TI-5050 and selling for $129.95 versus the $199.95 price tag of its predecessor. Dismantling the TI-5050M reveals a Single-Chip design based on the TMS1115, basically a TMS1000 with double the RAM cpacity and double the ROM capacity of the original TMS1000. Other examples of Multi-Chip designs with Digit processors include the TI-55-II introduced in 1981 with two TP0455 chips and the PC-800 Printer using a tandem of TMS1000 and TMS1300 Microcomputers.

Texas Instruments' ambitious LCD III Project proposed in January 1978 replacing all Register Processors used in their calculator portfolio with a family of three modular Digit Processors, differing mainly in ROM and RAM capacity and each chip optionally including additional modules for timekeeping functions and LC-Displays drivers. Well, the project didn't went well and the (failed) TP0485 was the last Digit Processor that Texas Instrument developed for Electronic Calculators, with a RAM capacity of 1,408 Bits and a ROM capacity of 41,800 Bits it was a fivefold improvement over the original TMS1000. Texas Instruments decided to work for future calculator chip designs with Toshiba and most of their calculators introduced in the 1980s use 4-bit Toshiba Microcomputer.

Here at the Datamath Calculator Museum we differentiate for calculators using Digit Processors five steps of evolution from the SR-16 based on the TMS1000:

The next major step in evolution was switching for Graphing Calculators to 8-bit Architectures with Processors like the Z80 and external RAM and ROM. The TI-81 introduced in 1990 is centered around a Toshiba T6A49A Application-specific 8-bit Z80 CPU, 128k Bytes Mask ROM, 8k Bytes Static RAM (SRAM) and three display drivers from Toshiba for the 64 * 96 pixel dot matrix LC-Display. This basic architecture branched into three product lines before being replaced in 1999 by the TI-83 Plus architecture using more flexible Flash ROM instead the Mask ROM:

Learn more about the Hardware Architecture of TI’s Graphing Calculators.

If you have additions to the above article please email: joerg@datamath.org.

© Joerg Woerner, September 9, 2023. No reprints without written permission.