It goes like this: a quantum of mantle is heated, rises, and as it does so, cooler material comes in from the side to take up the vacant space.  That's all.  That's all there is to it, ...buoyancy,  the basic driver.  However any child who has ever played with his or her rubber duck in the bath knows that this is only half the story.  The other half is  that the material coming in from the side is really (like water off a duck's back), coming from above, not from the side.  Plate tectonics however says that as the quantum of mantle ('balloon') rises, the whole column of mantle ahead of it is "dynamically forced" to rise.  The other end of this column at the Earth's surface is then "dynamically lifted" (in ridges), and that a process known as "ridge push" then sets in.  A lifted ridge is gravitationally unstable and so collapses away from the axis of uplift, 'pushing' down the slope. Pushing (or is it pulling?) in opposite directions creates a tension in the ridge, thus releasing the pressure immediately beneath the ridge and causing partial melting and intrusion of the melt into the growing dilating fracture.

So there we have it - solid rise in the mantle, 'dynamic forcing'/ 'lifting',  'plate accretion', and 'ridge push'  -  the growth of jargon to hide what is really considerable obfuscation.  Firstly there's the  'rubber duck' (water-slipping-off-a duck's- back) factor, which questions whether any column of mantle is ever lifted in the first place.  Secondly, why does this rising column find it necessary to lift the crust above gravitational equilibrium (instead of dissipating to accommodate gravity)?  Thirdly, why (or where) does this balloon find the energy to drive a double phase-change (solid - liquid - solid) at the ridge fractures?  And fourthly, why should the locus of this convectional  'shimmy'  remain fixed in space whilst the so-called 'cell' trundles along?

How many nonsenses are in there?  Four.  ...And so far we're not even half-way round, ...we haven't even got as far as that other nonsense, ...slab-pull.
 

Another nonsensical point that should be mentioned whilst we're at it, is to do with the age of the ocean floors.   In plate tectonics the age of the ocean floors depends heavily on the gradient of the slope away from the ridges as this reflects a measure of cooling and thermal contraction.   Note the delicate side-step.   The slope itself  is no longer related to ridge-push/ gravitational collapse, but to thermal contraction,  i.e., any gravitational collapse of this 'mountain belt' is just arbitrarily set aside whilst the notion of 'heat-as-driver' is reinstated. 
 

Or do we?   Transform faults virtually don't figure in a plate tectonic model, other than that they are (quote) "the mechanism by which one plate moves past another".   And why does a convection cell need transforms?  Well, the short answer is that it doesn't.  After all, pots of soup don't have transforms.  And what's more, transform faults are actually something of a substantial embarrassment to Plate Tectonics.

The whole lot is a convection Cell.  But they're there, embarassingly so, Lots of them, and so we need to take them into account (don't we?).   So what's to be done?  Well, the model obviously is ok,  it just  needs to be tweaked a bit.    Note, not discarded, and something else substituted that does take account of transforms  just 'tweaked'.   Just rack up the parameterisation factors  and Hey Presto! - transforms.  So, ..not zones of brittle failure, ..but ductile flow. 

Mechanism?   What mechanism?   You have to really peer to see transform faults on plate maps, so much so you'd hardly know they're there, so why bother with a mechanism at all?   The truth of this dismissive attitude to transforms is that there is virtually no convincing way in plate tectonics (or in the literature) how these develop (except the link above) - how and where the first transforms formed, how many might develop at once,  or if they are sequential, whether they are 'bottom-up' or 'top-down' structures, how they really partition 'plates', or why they are globally configured as a 'set' with rotational symmetry.   Despite the truly awesome aspect of ridge structure that they represent, transforms have been regarded with dismissive familiarity for decades.   They are however now an area of much research, but people don't seem to be getting anywhere fast,  and won't till the overall (obvious) gross spin symmetry of transforms is taken into account, as well as how they link to continental crustal deformation.  Answering the questions what transform faults really are,  how transform faults form and how they grow spells the demise of plate tectonics.   It's really a question of who's going to risk their career by belling the cat.  Thus far not a whisper.   But you can bet when there is one, it will be in the language of loud triumphalism, of the unbelievable phenomenal  success of plate tectonics.  That way everybody gets to stay clever, even the ones hanging off the back end of the wagon, who don't know whether it's turning right or left  - or even moving at all.  The ones getting barked at, by anklebiters like me.