Yet Another Discrete Computer with Silicon transistors

Similar projects worth following
It is meant to be the most advanced of my line of "very discrete" 16-bits computers : after the relay version ( inspired by the 40's) and the Germanium version ( for the 60's generation), I continue the historical exploration into the 80's with a cheaper, hence larger computer that follows the principles outlined by AMBAP ( )
Though the architecture is directly borrowed from the Relay and Ge projects, I'll explore a variation on the ECL theme, trying to use "non saturated logic", which departs from the RTL, DTL and TTL experiments of my fellows here.

I hope that 20K+ transistors will be enough.

As the steamroller of progress slowly moves forward, you can get a bag of 1K silicon transistors for under $20. Look at's listings, he has brand new, dirt-cheap BC549C, BC550C, BC559C... And I don't even look at the parts directly from the manufacturers ! It's only natural to move the BJT exploration to this brand new old technology. Compared to Germanium, silicon :

  • is way cheaper
  • leaks much less
  • has way better amplification (hFE > 400, vs 20-40 for the AF240)
  • comes in black plastic package

The BC55x are rated at "250MHz" only, vs 500MHz for the AF240, but the above advantages should greatly compensate ! (the hFE in particular)

"True ECL" is described in detail at I am studying circuits inspired by the ECL topology, but which are not totally conform (a bit cruder, easer to manage). "Non-Saturated Transistor Logic" is still a decent compromise for speed and consumption, compared to discrete DCTL/DTL/TTL.

For the faster, critical datapath circuits (adder, clock tree), I have found faster NPN transistors, such as the cheap 650MHz MPSH10 or the microwave-capable BFG425W or BFR96 The system speed target is set to 25MHz to ease synchronisation with a VGA display. We'll see if the whole BMOW can run fast enough...

At these speeds, computation is not black magic (at least I know what to expect), compared to memory which is the big roadblock. I'll cheat and use standard SRAM chips.

1. Inventory
2. Low-Voltage, Complementary Bipolar Logic
3. Faster and smaller

  • Faster and smaller

    Yann Guidon / YGDES03/26/2017 at 09:17 0 comments

    In the past months, I have been collecting ultra-fast bipolar transistors : BFR96, BF970, BFG425W, and an ample supply of MMBTH81.

    Lately I found an even faster reference, even cheaper, even more integrated :

    One of these SOT363 chips (like, SOT23 with 6 pins) can make a differential pair. 4 of them make a flip-flop.

    The BFS480 and the MMBTH81 make great parts for a fast, compact, bipolar processor...

  • Low-Voltage, Complementary Bipolar Logic

    Yann Guidon / YGDES01/10/2017 at 07:29 6 comments

    @Ted Yapo suggested a very unusual circuit in @esot.eric's thread

    Soon enough, he tried the circuit and created #CBJT Logic !

    I found out that Baker (of diode clamp fame) had explored this kind of circuit:

    And tonight, Ted tried some of the enhancements, which reduced the consumption and increased the speed !Speeding Up the NOT Gate

    The sweet spot seems to be between 1V and 1.1V with approximately 10ns of propagation time per inverter, and a not-too-high current draw...

    I am very tempted to play with this kind of gates for this project but experience with the other technologies show that a critical gate is the MUX2 (and MUX4) which are not trivial to design with this method. I should investigate "pass gates" made of NPN and PNP transistors...

  • Inventory

    Yann Guidon / YGDES01/06/2017 at 09:30 3 comments

    (20170326) (rev 20200524) The germanium inventory is there

    So far, my silicon stocks are:




    100 (?)

    Note: Ft (transition frequency) is often misleading:

    • For the KSP10 it's given as the "Current Gain Bandwidth Product", the part is tested up to 100MHz only (with some plots up to 1GHz).
    • The BFG425 is given as a "25GHz" part but the datasheet gives data tested up to 3GHz.

    Why a PNP in the middle of the NPN ? Because of Complementary ECL : it must be possible to "pack" some gates together to form a circuit with alternating polarities, save some time and gates, and power too. For example the BFS480 could be paired with the MMBTH81 (though the speed disparity might not be a great choice)

    I stumbled upon a stock of 2N2369, they might as well be included in this inventory "for scientific purposes". uses some of them and I tested the 5ns rise time with few efforts (it can go faster with more efforts).

    Note: some BC549C are already gone to contribute to #The Cardboard Computer :-)

    20200521 : added the PMBT2369 and the BFP420

View all 3 project logs

Enjoy this project?



esot.eric wrote 03/31/2018 at 03:08 point

Hah! "Silicon?! I never thought I'd see the day..." Then scrolling down I see a very familiar and yet, quite obscure topology... :)

  Are you sure? yes | no

Yann Guidon / YGDES wrote 03/31/2018 at 03:13 point


I haven't sorted out all the parameters because there are so many possible topologies for TTL / DTL / DCTL / RTL / ECL circuits.... I'm still researching and I've got a few helpful old books :-D

  Are you sure? yes | no

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates